I have tried out two different Aikido dojos this week. Emerald City Aikido and Puget Sound Aikikai (I live in Seattle WA). Both differ from one another a fair bit in terms of the \"feel\" of the classroom environment and instruction. Emerald City Aikido was much more casual and had a comfortable enough environment that I had a good dialog forming between myself and the sensei. The other dojo was much more intense. It and its students seemed much more focused and serious and I never really felt comfortable enough with the teacher to really discuss my form or the techniques. Also Puget Sound Aikikai is affiliated with the USAF whereas the other is an independent dojo.
At this point I'm heavily confused. Both seem great but in different ways. Any suggestions?"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Ask HN: Choosing an Aikido Dojo (Seattle)"},"url":{"matchLevel":"none","matchedWords":[],"value":""}},"_tags":["story","author_darkxanthos","story_1892844","ask_hn"],"author":"darkxanthos","children":[1892866,1892867],"created_at":"2010-11-11T04:04:57Z","created_at_i":1289448297,"num_comments":3,"objectID":"1892844","points":2,"story_id":1892844,"story_text":"I am learning Aikido for exercise, self-defense, and as a study in philosophy.
I have tried out two different Aikido dojos this week. Emerald City Aikido and Puget Sound Aikikai (I live in Seattle WA). Both differ from one another a fair bit in terms of the \"feel\" of the classroom environment and instruction. Emerald City Aikido was much more casual and had a comfortable enough environment that I had a good dialog forming between myself and the sensei. The other dojo was much more intense. It and its students seemed much more focused and serious and I never really felt comfortable enough with the teacher to really discuss my form or the techniques. Also Puget Sound Aikikai is affiliated with the USAF whereas the other is an independent dojo.
At this point I'm heavily confused. Both seem great but in different ways. Any suggestions?","title":"Ask HN: Choosing an Aikido Dojo (Seattle)","updated_at":"2024-09-19T17:24:36Z","url":""},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"booffa"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Meowth GBA Translator is an open-source, AI-powered tool that automates translation of Pok\u00e9mon GBA ROMs (including binary hacks like FireRed, Emerald, Ruby/Sapphire, and Mystery Dungeon). Powered by LLMs (supports OpenAI, DeepSeek, Gemini, Claude, Groq, and 10+ others), it extracts text, translates intelligently while preserving codes and context, then rebuilds the ROM \u2014 all in one click via a friendly GUI or simple CLI command.\nSupports 6+ languages (Chinese, English, French, German, Italian, Spanish) with optimized prompts and smart font patching. Focus on gameplay mods, let AI handle the words. Free, MIT-licensed, cross-platform."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: AI-powered one-click translator for Pok\u00e9mon GBA ROM hacks"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/Olcmyk/Meowth-GBA-Translator"}},"_tags":["story","author_booffa","story_47347680","show_hn"],"author":"booffa","children":[47347747,47347798,47348943,47385420],"created_at":"2026-03-12T07:45:32Z","created_at_i":1773301532,"num_comments":4,"objectID":"47347680","points":4,"story_id":47347680,"story_text":"Meowth GBA Translator is an open-source, AI-powered tool that automates translation of Pok\u00e9mon GBA ROMs (including binary hacks like FireRed, Emerald, Ruby/Sapphire, and Mystery Dungeon). Powered by LLMs (supports OpenAI, DeepSeek, Gemini, Claude, Groq, and 10+ others), it extracts text, translates intelligently while preserving codes and context, then rebuilds the ROM \u2014 all in one click via a friendly GUI or simple CLI command.\nSupports 6+ languages (Chinese, English, French, German, Italian, Spanish) with optimized prompts and smart font patching. Focus on gameplay mods, let AI handle the words. Free, MIT-licensed, cross-platform.","title":"Show HN: AI-powered one-click translator for Pok\u00e9mon GBA ROM hacks","updated_at":"2026-03-20T08:08:09Z","url":"https://github.com/Olcmyk/Meowth-GBA-Translator"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kmgrassi"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hey HN folks!
Emerald founder here - I\u2019m a dev, pediatrician and father. I\u2019ve been thinking about this idea for a long time. I recently sold my last company [0] and this seems like a good time to try something a little more risky but that I\u2019m passionate about.
In the distant past most kids worked on farms from a young age (I live in upstate NY and families with farms still have their kids work from a young age). With industrialization we got away from child labor because children were being exploited and exposed to dangerous factory environments.
With Emerald, I\u2019d like to challenge the notion that \u201cchild labor\u201d is necessarily a bad thing. As a pediatrician and parent I can tell you that American children could benefit from a dose of the real world. There are many important benefits to having the responsibilities of a job at a young age. I outlined several of these below.
The idea behind Emerald is to use the loophole that kids can work for their parents. We are going to create LLCs and bank accounts (will probably use Stipe Atlas and Mercury bank) for each family and then contract work through that LLC. This way kids can work on projects, earn money and gain experience.
Emerald will be a marketplace similar to Fiverr or Upwork. Companies or people who need work done can post the spec and then will be matched with a workforce who can complete the work.
The main difference will be that Emerald will be much more hands on during the work completion. Instead of individual freelancers bidding for work Emerald will construct teams of youth (prob 2-5 per team - maybe more?) with an adult mentor. The mentor will be paid and will oversee the work. The youth team will then collaborate to complete work. Lots of unknowns here and we will have to iterate on the best process.
There will also be extensive safety tooling and oversight. We plan to require all conversations to occur on-platform (Slack like chat), set time limits and give parents dashboards/controls.
The obvious first type of work to be done are things that teens are good at - content editing, content creation and software dev. Although I\u2019d like to hear other ideas for work that teens might be well suited for.
Benefits:\n- Earn money\n- Learn time management\n- Learn how to work with a team\n- Learn real world software development or tech skills\n- Start creating a resume and list of references
Risks:\n- Time commitment \n- Increase screen time. Kids already have too much screen time\n- Shorten childhood - we should let \u201ckids be kids\u201d for longer\n- Exploitation - companies taking advantage of inexperienced workers (paying less, etc)\n- Cause more work for parents (monitoring, etc)\n- Parents masquerading as their kids and doing the work (I don\u2019t know why they\u2019d do this but figure I\u2019d list it)
Questions:\n- What are some additional risks?\n- What are additional benefits?\n- Are there examples of programs like this out there? If so, what are the learnings?
Side note on AI\nI\u2019d like to avoid the \u201cAI will take all these jobs anyway\u201d rabbit hole. I\u2019d like to assume we can find something productive for kids (or any of us) to do.
Who I\u2019m looking for\n- Educators or people who have worked with teens on longer-term group projects - learnings from these projects\n- People interested in being mentors for teens on Emerald - will be paid positions\n- People with children who are interested in the initial pilot - no dates or commitment are set yet. Just if you are remotely interested.\n- Companies/people who are interested in submitting projects that they need completed. We can help you scope the project so it can be amorphous.
This is VERY early for Emerald. We haven\u2019t built any tech yet (aside from the very basic landing page). If anyone has interest in this idea feel free to reach out - my HN handle at gmail.
Thanks for reading this far!
[0]: PlentiAI acquired by Digiphy - https://www.digiphy.it/"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"Emerald \u2013 Work marketplace for teens/youth"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"https://www.emerald.new/"}},"_tags":["story","author_kmgrassi","story_42844176"],"author":"kmgrassi","children":[42844187,42844190,42844476],"created_at":"2025-01-27T18:35:50Z","created_at_i":1738002950,"num_comments":2,"objectID":"42844176","points":2,"story_id":42844176,"story_text":"Hey HN folks!
Emerald founder here - I\u2019m a dev, pediatrician and father. I\u2019ve been thinking about this idea for a long time. I recently sold my last company [0] and this seems like a good time to try something a little more risky but that I\u2019m passionate about.
In the distant past most kids worked on farms from a young age (I live in upstate NY and families with farms still have their kids work from a young age). With industrialization we got away from child labor because children were being exploited and exposed to dangerous factory environments.
With Emerald, I\u2019d like to challenge the notion that \u201cchild labor\u201d is necessarily a bad thing. As a pediatrician and parent I can tell you that American children could benefit from a dose of the real world. There are many important benefits to having the responsibilities of a job at a young age. I outlined several of these below.
The idea behind Emerald is to use the loophole that kids can work for their parents. We are going to create LLCs and bank accounts (will probably use Stipe Atlas and Mercury bank) for each family and then contract work through that LLC. This way kids can work on projects, earn money and gain experience.
Emerald will be a marketplace similar to Fiverr or Upwork. Companies or people who need work done can post the spec and then will be matched with a workforce who can complete the work.
The main difference will be that Emerald will be much more hands on during the work completion. Instead of individual freelancers bidding for work Emerald will construct teams of youth (prob 2-5 per team - maybe more?) with an adult mentor. The mentor will be paid and will oversee the work. The youth team will then collaborate to complete work. Lots of unknowns here and we will have to iterate on the best process.
There will also be extensive safety tooling and oversight. We plan to require all conversations to occur on-platform (Slack like chat), set time limits and give parents dashboards/controls.
The obvious first type of work to be done are things that teens are good at - content editing, content creation and software dev. Although I\u2019d like to hear other ideas for work that teens might be well suited for.
Benefits:\n- Earn money\n- Learn time management\n- Learn how to work with a team\n- Learn real world software development or tech skills\n- Start creating a resume and list of references
Risks:\n- Time commitment \n- Increase screen time. Kids already have too much screen time\n- Shorten childhood - we should let \u201ckids be kids\u201d for longer\n- Exploitation - companies taking advantage of inexperienced workers (paying less, etc)\n- Cause more work for parents (monitoring, etc)\n- Parents masquerading as their kids and doing the work (I don\u2019t know why they\u2019d do this but figure I\u2019d list it)
Questions:\n- What are some additional risks?\n- What are additional benefits?\n- Are there examples of programs like this out there? If so, what are the learnings?
Side note on AI\nI\u2019d like to avoid the \u201cAI will take all these jobs anyway\u201d rabbit hole. I\u2019d like to assume we can find something productive for kids (or any of us) to do.
Who I\u2019m looking for\n- Educators or people who have worked with teens on longer-term group projects - learnings from these projects\n- People interested in being mentors for teens on Emerald - will be paid positions\n- People with children who are interested in the initial pilot - no dates or commitment are set yet. Just if you are remotely interested.\n- Companies/people who are interested in submitting projects that they need completed. We can help you scope the project so it can be amorphous.
This is VERY early for Emerald. We haven\u2019t built any tech yet (aside from the very basic landing page). If anyone has interest in this idea feel free to reach out - my HN handle at gmail.
Thanks for reading this far!
[0]: PlentiAI acquired by Digiphy - https://www.digiphy.it/","title":"Emerald \u2013 Work marketplace for teens/youth","updated_at":"2025-01-28T14:40:00Z","url":"https://www.emerald.new/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ajcarpy2005"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"New Futurist Magazine to Teach Implementable AI"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"https://www.kickstarter.com/projects/coreylemont/the-emerald-ethos"}},"_tags":["story","author_ajcarpy2005","story_20099945"],"author":"ajcarpy2005","created_at":"2019-06-04T22:39:01Z","created_at_i":1559687941,"num_comments":0,"objectID":"20099945","points":1,"story_id":20099945,"title":"New Futurist Magazine to Teach Implementable AI","updated_at":"2024-09-20T04:21:52Z","url":"https://www.kickstarter.com/projects/coreylemont/the-emerald-ethos"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"adia-dev"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Pretty self-explanatory tbh.\nI kept hitting the same friction using LeetCode, the built-in editor has no real language support and none of the Vim muscle memory I rely on everywhere else. So I started building it in.
This is also where I love AI, I had started to work on it years ago, at first it was only snippets of most used classes (vector, string, etc...) and a few vim keymaps.\nBeing able to tell Codex "Improve this, make no mistakes" is pretty fun.
Also side note, I've always used Leetcode as a game to practice problem solving (Graph Algorithms <3333), but I feel like now more than ever it is a good platform to code manually and challenge yourself, especially when vibe coding is so present. \n(I don't recommend grinding if you don't enjoy it tho)"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: C++ LSP for the LeetCode editor and Vim keybindings"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"https://twitter.com/emerald_d3v/status/2106885158360195426"}},"_tags":["story","author_adia-dev","story_49958951","show_hn"],"author":"adia-dev","children":[49959594],"created_at":"2026-10-04T23:18:31Z","created_at_i":1791155911,"num_comments":0,"objectID":"49958951","points":1,"story_id":49958951,"story_text":"Pretty self-explanatory tbh.\nI kept hitting the same friction using LeetCode, the built-in editor has no real language support and none of the Vim muscle memory I rely on everywhere else. So I started building it in.
This is also where I love AI, I had started to work on it years ago, at first it was only snippets of most used classes (vector, string, etc...) and a few vim keymaps.\nBeing able to tell Codex "Improve this, make no mistakes" is pretty fun.
Also side note, I've always used Leetcode as a game to practice problem solving (Graph Algorithms <3333), but I feel like now more than ever it is a good platform to code manually and challenge yourself, especially when vibe coding is so present. \n(I don't recommend grinding if you don't enjoy it tho)","title":"Show HN: C++ LSP for the LeetCode editor and Vim keybindings","updated_at":"2026-10-05T01:02:26Z","url":"https://twitter.com/emerald_d3v/status/2106885158360195426"},{"_highlightResult":{"author":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"Emerald_dreamer"},"title":{"matchLevel":"none","matchedWords":[],"value":"Humanity Is Coming to an End"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://dearworld.ai/"}},"_tags":["story","author_Emerald_dreamer","story_45578018"],"author":"Emerald_dreamer","children":[45578047,45578050,45578081,45578134,45578155,45578184,45578196,45578212,45578331,45578392,45578553,45578801,45579583],"created_at":"2025-10-14T09:41:44Z","created_at_i":1760434904,"num_comments":9,"objectID":"45578018","points":12,"story_id":45578018,"title":"Humanity Is Coming to an End","updated_at":"2026-03-05T22:52:14Z","url":"https://dearworld.ai/"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"partial","matchedWords":["emerald"],"value":"emeraldd"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"AI company has released an app that lets people converse with avatars of dead"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://old.reddit.com/r/STEW_ScTecEngWorld/comments/1owliqk/an_ai_company_has_released_an_app_that_lets/"}},"_tags":["story","author_emeraldd","story_46440024"],"author":"emeraldd","children":[46440039,46440077,46440236],"created_at":"2025-12-31T00:40:56Z","created_at_i":1767141656,"num_comments":3,"objectID":"46440024","points":2,"story_id":46440024,"title":"AI company has released an app that lets people converse with avatars of dead","updated_at":"2026-03-05T23:16:28Z","url":"https://old.reddit.com/r/STEW_ScTecEngWorld/comments/1owliqk/an_ai_company_has_released_an_app_that_lets/"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"partial","matchedWords":["emerald"],"value":"emeraldd"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Aircraft Carrier USS Enterprise to Be Decommissioned"},"url":{"matchLevel":"none","matchedWords":[],"value":"http://www.forces.tv/05981801"}},"_tags":["story","author_emeraldd","story_13494855"],"author":"emeraldd","children":[13494858],"created_at":"2017-01-26T20:11:39Z","created_at_i":1485461499,"num_comments":1,"objectID":"13494855","points":2,"story_id":13494855,"title":"Aircraft Carrier USS Enterprise to Be Decommissioned","updated_at":"2024-09-20T00:20:45Z","url":"http://www.forces.tv/05981801"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"partial","matchedWords":["emerald"],"value":"emeraldd"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Illustrations of Madness: James Tilly Matthews and the Air Loom"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://publicdomainreview.org/2014/11/12/illustrations-of-madness-james-tilly-matthews-and-the-air-loom/"}},"_tags":["story","author_emeraldd","story_12774236"],"author":"emeraldd","created_at":"2016-10-23T16:43:38Z","created_at_i":1477241018,"num_comments":0,"objectID":"12774236","points":1,"story_id":12774236,"title":"Illustrations of Madness: James Tilly Matthews and the Air Loom","updated_at":"2024-09-19T23:54:03Z","url":"https://publicdomainreview.org/2014/11/12/illustrations-of-madness-james-tilly-matthews-and-the-air-loom/"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"partial","matchedWords":["emerald"],"value":"emeraldd"},"title":{"matchLevel":"none","matchedWords":[],"value":"California bans ITT tech from accepting new students"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"http://www.latimes.com/business/la-fi-itt-tech-aid-20160826-snap-story.html?utm_source=fark&utm_medium=website&utm_content=link&ICID=ref_fark"}},"_tags":["story","author_emeraldd","story_12386777"],"author":"emeraldd","children":[12386940,12386941,12386945,12387005,12387006,12387025,12387034,12387132,12387145,12387177,12387182,12387221,12387298,12387329,12387355,12387387,12387389,12387420,12387485,12387546,12387617,12387637,12387702,12387723,12387945,12387982,12388040,12388246,12388339,12388359,12389642,12390480],"created_at":"2016-08-30T01:30:27Z","created_at_i":1472520627,"num_comments":312,"objectID":"12386777","points":293,"story_id":12386777,"title":"California bans ITT tech from accepting new students","updated_at":"2026-09-18T04:54:08Z","url":"http://www.latimes.com/business/la-fi-itt-tech-aid-20160826-snap-story.html?utm_source=fark&utm_medium=website&utm_content=link&ICID=ref_fark"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"akshaysg"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hi HN! We\u2019re Akshay and Jake. We put together a tool called Haystack to make pull requests straightforward to read.
What Haystack does:
-- Builds a clear narrative. Changes in Haystack aren\u2019t just arranged as unordered diffs. Instead, they unfold in a logical order, each paired with an explanation in plain, precise language
-- Focuses attention where it counts. Routine plumbing and refactors are put into skimmable sections so you can spend your time on design and correctness
-- Provides full cross-file context. Every new or changed function/variable is traced across the codebase, showing how it\u2019s used beyond the immediate diff
Here\u2019s a quick demo: https://youtu.be/w5Lq5wBUS-I
If you\u2019d like to give it a spin, head over to haystackeditor.com/review! We set up some demo PRs that you should be able to understand and review even if you\u2019ve never seen the repos before!
We used to work at big companies, where reviewing non-trivial pull requests felt like reading a book with its pages out of order. We would jump and scroll between files, trying to piece together the author\u2019s intent before we could even start reviewing. And, as authors, we would spend time to restructure our own commits just to make them readable.\nAI has made this even trickier. Today it\u2019s not uncommon for a pull request to contain code the author doesn\u2019t fully understand themselves!
So, we built Haystack to help reviewers spend less time untangling code and more time giving meaningful feedback. We would love to hear about whether it gets the job done for you!
How we got here:
Haystack began as (yet another) VS Code fork where we experimented with visualizing code changes on a canvas. At first, it was a neat way to show how pieces of code worked together. But customers started laying out their entire codebase just to make sense of it. That\u2019s when we realized the deeper problem: understanding a codebase is hard, and engineers need better ways to quickly understand unfamiliar code.
As we kept building, another insight emerged: with AI woven into workflows, engineers don\u2019t always need to master every corner of a codebase to ship features. But in code review, deep and continuous context still matters, especially to separate what\u2019s important to review from plumbing and follow-on changes.
So we pivoted. We took what we\u2019d learned and worked closely with engineers to refine the idea. We started with simple code analysis (using language servers, tree-sitter, etc.) to show how changes relate. Then we added AI to explain and organize those changes and to trace how data moves through a pull request. Finally, we fused the two by empowering AI agents to use static analyses. Step by step, that became the Haystack we\u2019re showing today.
We\u2019d love to hear your thoughts, feedback, or suggestions!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Haystack \u2013 Review pull requests like you wrote them yourself"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://haystackeditor.com"}},"_tags":["story","author_akshaysg","story_45201703","show_hn"],"author":"akshaysg","children":[45202592,45202668,45202692,45202881,45203343,45203664,45203809,45204406,45204414,45204477,45204868,45206297,45206564,45206861,45207872,45209952,45209978,45210506,45212937],"created_at":"2025-09-10T18:21:12Z","created_at_i":1757528472,"num_comments":59,"objectID":"45201703","points":88,"story_id":45201703,"story_text":"Hi HN! We\u2019re Akshay and Jake. We put together a tool called Haystack to make pull requests straightforward to read.
What Haystack does:
-- Builds a clear narrative. Changes in Haystack aren\u2019t just arranged as unordered diffs. Instead, they unfold in a logical order, each paired with an explanation in plain, precise language
-- Focuses attention where it counts. Routine plumbing and refactors are put into skimmable sections so you can spend your time on design and correctness
-- Provides full cross-file context. Every new or changed function/variable is traced across the codebase, showing how it\u2019s used beyond the immediate diff
Here\u2019s a quick demo: https://youtu.be/w5Lq5wBUS-I
If you\u2019d like to give it a spin, head over to haystackeditor.com/review! We set up some demo PRs that you should be able to understand and review even if you\u2019ve never seen the repos before!
We used to work at big companies, where reviewing non-trivial pull requests felt like reading a book with its pages out of order. We would jump and scroll between files, trying to piece together the author\u2019s intent before we could even start reviewing. And, as authors, we would spend time to restructure our own commits just to make them readable.\nAI has made this even trickier. Today it\u2019s not uncommon for a pull request to contain code the author doesn\u2019t fully understand themselves!
So, we built Haystack to help reviewers spend less time untangling code and more time giving meaningful feedback. We would love to hear about whether it gets the job done for you!
How we got here:
Haystack began as (yet another) VS Code fork where we experimented with visualizing code changes on a canvas. At first, it was a neat way to show how pieces of code worked together. But customers started laying out their entire codebase just to make sense of it. That\u2019s when we realized the deeper problem: understanding a codebase is hard, and engineers need better ways to quickly understand unfamiliar code.
As we kept building, another insight emerged: with AI woven into workflows, engineers don\u2019t always need to master every corner of a codebase to ship features. But in code review, deep and continuous context still matters, especially to separate what\u2019s important to review from plumbing and follow-on changes.
So we pivoted. We took what we\u2019d learned and worked closely with engineers to refine the idea. We started with simple code analysis (using language servers, tree-sitter, etc.) to show how changes relate. Then we added AI to explain and organize those changes and to trace how data moves through a pull request. Finally, we fused the two by empowering AI agents to use static analyses. Step by step, that became the Haystack we\u2019re showing today.
We\u2019d love to hear your thoughts, feedback, or suggestions!","title":"Show HN: Haystack \u2013 Review pull requests like you wrote them yourself","updated_at":"2026-05-18T23:28:41Z","url":"https://haystackeditor.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"SencerH"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"There was a time when most of us kept archives of the digital media we loved. FTP servers, Usenet newsgroups, followed by Napster, eDonkey, and torrent networks were our primary archival sources.
Gigabytes of hard drives kept filling up to the brim. We would stare longingly at movies and music albums we couldn't acquire simply because we ran out of disk space. We spent hours agonizing over what to delete to free up space or figuring out how to compress files more efficiently.\nThen they introduced cloud storage, which has now become the backbone of most modern services.
At first, we started uploading our personal files to these services. Then media streaming services began sprouting up everywhere. Slowly, we deleted the movies and music we actually owned. We sold our hard drives. We threw away the CD archives we had spent hours burning. And just like that, we became able to access everything from anywhere.
Then AI emerged and rapidly became incredibly capable. At first, we used it just to ask questions and get answers. Then we realized we could generate music and videos with it.
Before long, AI-generated music started flooding streaming platforms.\nNow, when I search for Guns N' Roses on YouTube to listen to, instead of the original, I get a bunch of AI-generated mixes featuring them and similar genres.
You open something up to listen, and sure, maybe a G'N'R track plays, but it is an AI-remade version. The songs that follow are the same. Yes, the rhythm might be identical, but what I feel is no different than drinking a glass of soda that has gone completely flat.
And I realized that accessing original content is getting harder by the day.
If things keep going this way, reaching original content will soon become nearly impossible. This feels even more alarming when you learn how aggressively AI companies scan and digitize original books, treat their physical forms like squeezed-out orange peels, and then discard or destroy them\u2014making me understand the true value of archiving more than ever.
And today, I feel luckier every single day for still holding onto my MP3 archive."},"title":{"matchLevel":"none","matchedWords":[],"value":"The Importance of Personal Archiving"}},"_tags":["story","author_SencerH","story_49552636","ask_hn"],"author":"SencerH","created_at":"2026-09-03T16:25:03Z","created_at_i":1788452703,"num_comments":0,"objectID":"49552636","points":4,"story_id":49552636,"story_text":"There was a time when most of us kept archives of the digital media we loved. FTP servers, Usenet newsgroups, followed by Napster, eDonkey, and torrent networks were our primary archival sources.
Gigabytes of hard drives kept filling up to the brim. We would stare longingly at movies and music albums we couldn't acquire simply because we ran out of disk space. We spent hours agonizing over what to delete to free up space or figuring out how to compress files more efficiently.\nThen they introduced cloud storage, which has now become the backbone of most modern services.
At first, we started uploading our personal files to these services. Then media streaming services began sprouting up everywhere. Slowly, we deleted the movies and music we actually owned. We sold our hard drives. We threw away the CD archives we had spent hours burning. And just like that, we became able to access everything from anywhere.
Then AI emerged and rapidly became incredibly capable. At first, we used it just to ask questions and get answers. Then we realized we could generate music and videos with it.
Before long, AI-generated music started flooding streaming platforms.\nNow, when I search for Guns N' Roses on YouTube to listen to, instead of the original, I get a bunch of AI-generated mixes featuring them and similar genres.
You open something up to listen, and sure, maybe a G'N'R track plays, but it is an AI-remade version. The songs that follow are the same. Yes, the rhythm might be identical, but what I feel is no different than drinking a glass of soda that has gone completely flat.
And I realized that accessing original content is getting harder by the day.
If things keep going this way, reaching original content will soon become nearly impossible. This feels even more alarming when you learn how aggressively AI companies scan and digitize original books, treat their physical forms like squeezed-out orange peels, and then discard or destroy them\u2014making me understand the true value of archiving more than ever.
And today, I feel luckier every single day for still holding onto my MP3 archive.","title":"The Importance of Personal Archiving","updated_at":"2026-09-08T16:50:17Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"RandomHooman"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"2001-2 tech bust - Amazon, Google, SFDC emerged \n2008-9 crisis - Uber, AirBnB emerged
Given the current market, time is ripe for another next-gen company to emerge.
How to spot it :-)?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Spot the Next FAANG?"}},"_tags":["story","author_RandomHooman","story_38036068","ask_hn"],"author":"RandomHooman","children":[38036128,38036155,38036340],"created_at":"2023-10-27T08:42:28Z","created_at_i":1698396148,"num_comments":2,"objectID":"38036068","points":2,"story_id":38036068,"story_text":"2001-2 tech bust - Amazon, Google, SFDC emerged \n2008-9 crisis - Uber, AirBnB emerged
Given the current market, time is ripe for another next-gen company to emerge.
How to spot it :-)?","title":"Spot the Next FAANG?","updated_at":"2024-09-20T15:27:10Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jotbot"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Anyone have any experience with these container orchestration platforms?"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Ask HN: Thoughts on HPE Ezmeral/Private Cloud AI"}},"_tags":["story","author_jotbot","story_44499851","ask_hn"],"author":"jotbot","created_at":"2025-07-08T13:38:18Z","created_at_i":1751981898,"num_comments":0,"objectID":"44499851","points":1,"story_id":44499851,"story_text":"Anyone have any experience with these container orchestration platforms?","title":"Ask HN: Thoughts on HPE Ezmeral/Private Cloud AI","updated_at":"2025-07-08T13:43:14Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jrepinc"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"AI\u2019s algorithmic violence emerged from our own social matrix"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://thesociologicalreview.org/magazine/june-2023/artificial-intelligence/predicted-benefits-proven-harms/"}},"_tags":["story","author_jrepinc","story_36216011"],"author":"jrepinc","created_at":"2023-06-06T17:13:49Z","created_at_i":1686071629,"num_comments":0,"objectID":"36216011","points":2,"story_id":36216011,"title":"AI\u2019s algorithmic violence emerged from our own social matrix","updated_at":"2024-09-20T14:11:05Z","url":"https://thesociologicalreview.org/magazine/june-2023/artificial-intelligence/predicted-benefits-proven-harms/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rickbeaver"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"<p><strong>I. The Rise of AI Writing Tools</strong></p>\n<p>AI writing tools, such as the revolutionary Essay Generator from <a href="https://www.perfectessaywriter.ai/" rel="DoFollow">perfectessaywriter.ai</a>, have emerged as game-changers in the writing landscape. These tools leverage advanced algorithms and natural language processing to assist writers in various aspects of the writing process, from generating ideas and creating outlines to crafting well-structured and coherent essays.</p>\n<p style="text-align: center;"><img src="https://custom-images.strikinglycdn.com/res/hrscywv4p/image/upload/c_limit,fl_lossy,h_9000,w_1200,f_auto,q_auto/9519882/61226_858012.png" alt="" /></p>\n<p><strong>1. Press Releases and News Mentions of PerfectEssayWriterAI</strong></p>\n<p>PerfectEssayWriterAI has garnered significant attention in the press, with notable mentions in press releases and news articles. These include publications like <a href="https://www.startupguys.net/best-ai-essay-generator-tools/" target="_new">StartupGuys</a>, <a href="https://markets.businessinsider.com/news/stocks/perfectessaywriter-ai-introduces-the-easy-ai-essay-writing-tool-of-2023-1032351992" target="_new">Markets Insider</a>, <a href="https://www.ilounge.com/articles/best-5-ai-essay-writing-tools-online-review-2023-24" target="_new">iLounge</a>, and <a href="https://goodmenproject.com/technology/top-5-ai-essay-writing-tools-of-2023/" target="_new">Good Men Project</a>, showcasing the recognition and impact of PerfectEssayWriterAI's AI writing tool in the industry. </p>\n<h3>Crafting a Strong Foundation</h3>\n<p>Selecting appropriate <a href="https://www.perfectessaywriter.ai/how-to-write-an-essay/essay-topics" rel="DoFollow">Essay Topics</a> is the first step in creating a compelling and engaging piece of writing. Once you have a topic in mind, it is crucial to establish a well-organized essay structure to effectively convey your ideas. An <a href="https://www.perfectessaywriter.ai/how-to-write-an-essay/essay-structure" rel="DoFollow">Essay Structure</a> typically consists of an introduction, body paragraphs, and a conclusion. Furthermore, creating an <a href="https://www.perfectessaywriter.ai/how-to-write-an-essay/essay-outline" rel="DoFollow">Essay Outline</a> can help you plan and organize your thoughts before diving into the writing process. Outlining provides a roadmap for your essay, ensuring that your arguments flow logically and coherently from one paragraph to another.</p>\n<p>In this dynamic landscape, PerfectEssayWriterAI's AI writing tool and WriteMyessayHelp's custom essay writing services stand out as exemplary providers. PerfectEssayWriterAI's <a href="https://www.perfectessaywriter.ai/" rel="DoFollow">Essay generator</a> empowers writers with efficient content generation, while WriteMyessayHelp's expertise ensures personalized and high-quality essays.</p>\n<h2><strong>I</strong><strong>II</strong><strong>. The Future of custom writing services</strong></h2>\n<p>As AI technology continues to advance, custom writing services will continue to adapt and evolve. The integration of AI <a href="https://medium.com/ai-essay-writer">AI essay writer</a> tools and custom writing services will become even more seamless, resulting in an enhanced writing experience for writers of all backgrounds.</p>"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Evolution of Custom Writing Services Adapting to AI Writing Tools and Technology"}},"_tags":["story","author_rickbeaver","story_36664330","ask_hn"],"author":"rickbeaver","created_at":"2023-07-10T10:29:11Z","created_at_i":1688984951,"num_comments":0,"objectID":"36664330","points":1,"story_id":36664330,"story_text":"<p><strong>I. The Rise of AI Writing Tools</strong></p>\n<p>AI writing tools, such as the revolutionary Essay Generator from <a href="https://www.perfectessaywriter.ai/" rel="DoFollow">perfectessaywriter.ai</a>, have emerged as game-changers in the writing landscape. These tools leverage advanced algorithms and natural language processing to assist writers in various aspects of the writing process, from generating ideas and creating outlines to crafting well-structured and coherent essays.</p>\n<p style="text-align: center;"><img src="https://custom-images.strikinglycdn.com/res/hrscywv4p/image/upload/c_limit,fl_lossy,h_9000,w_1200,f_auto,q_auto/9519882/61226_858012.png" alt="" /></p>\n<p><strong>1. Press Releases and News Mentions of PerfectEssayWriterAI</strong></p>\n<p>PerfectEssayWriterAI has garnered significant attention in the press, with notable mentions in press releases and news articles. These include publications like <a href="https://www.startupguys.net/best-ai-essay-generator-tools/" target="_new">StartupGuys</a>, <a href="https://markets.businessinsider.com/news/stocks/perfectessaywriter-ai-introduces-the-easy-ai-essay-writing-tool-of-2023-1032351992" target="_new">Markets Insider</a>, <a href="https://www.ilounge.com/articles/best-5-ai-essay-writing-tools-online-review-2023-24" target="_new">iLounge</a>, and <a href="https://goodmenproject.com/technology/top-5-ai-essay-writing-tools-of-2023/" target="_new">Good Men Project</a>, showcasing the recognition and impact of PerfectEssayWriterAI's AI writing tool in the industry. </p>\n<h3>Crafting a Strong Foundation</h3>\n<p>Selecting appropriate <a href="https://www.perfectessaywriter.ai/how-to-write-an-essay/essay-topics" rel="DoFollow">Essay Topics</a> is the first step in creating a compelling and engaging piece of writing. Once you have a topic in mind, it is crucial to establish a well-organized essay structure to effectively convey your ideas. An <a href="https://www.perfectessaywriter.ai/how-to-write-an-essay/essay-structure" rel="DoFollow">Essay Structure</a> typically consists of an introduction, body paragraphs, and a conclusion. Furthermore, creating an <a href="https://www.perfectessaywriter.ai/how-to-write-an-essay/essay-outline" rel="DoFollow">Essay Outline</a> can help you plan and organize your thoughts before diving into the writing process. Outlining provides a roadmap for your essay, ensuring that your arguments flow logically and coherently from one paragraph to another.</p>\n<p>In this dynamic landscape, PerfectEssayWriterAI's AI writing tool and WriteMyessayHelp's custom essay writing services stand out as exemplary providers. PerfectEssayWriterAI's <a href="https://www.perfectessaywriter.ai/" rel="DoFollow">Essay generator</a> empowers writers with efficient content generation, while WriteMyessayHelp's expertise ensures personalized and high-quality essays.</p>\n<h2><strong>I</strong><strong>II</strong><strong>. The Future of custom writing services</strong></h2>\n<p>As AI technology continues to advance, custom writing services will continue to adapt and evolve. The integration of AI <a href="https://medium.com/ai-essay-writer">AI essay writer</a> tools and custom writing services will become even more seamless, resulting in an enhanced writing experience for writers of all backgrounds.</p>","title":"Evolution of Custom Writing Services Adapting to AI Writing Tools and Technology","updated_at":"2024-09-20T14:37:19Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"bko"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Pangram Has Emerged as the Gold Standard of AI Detection. Should You Trust It?"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"https://www.wired.com/story/pangram-has-emerged-as-the-gold-standard-of-ai-detection/"}},"_tags":["story","author_bko","story_49558711"],"author":"bko","children":[49558955,49574556],"created_at":"2026-09-03T23:51:21Z","created_at_i":1788479481,"num_comments":3,"objectID":"49558711","points":5,"story_id":49558711,"title":"Pangram Has Emerged as the Gold Standard of AI Detection. Should You Trust It?","updated_at":"2026-09-06T04:07:39Z","url":"https://www.wired.com/story/pangram-has-emerged-as-the-gold-standard-of-ai-detection/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Brajeshwar"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Pangram Has Emerged as the Gold Standard of AI Detection"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"https://www.wired.com/story/pangram-has-emerged-as-the-gold-standard-of-ai-detection/"}},"_tags":["story","author_Brajeshwar","story_49538687"],"author":"Brajeshwar","children":[49539225],"created_at":"2026-09-02T16:28:46Z","created_at_i":1788366526,"num_comments":0,"objectID":"49538687","points":5,"story_id":49538687,"title":"Pangram Has Emerged as the Gold Standard of AI Detection","updated_at":"2026-09-03T06:32:30Z","url":"https://www.wired.com/story/pangram-has-emerged-as-the-gold-standard-of-ai-detection/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"seeksky"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Realizing a one-person company and developing multi-agent collaborative applications for multiple projects in parallel, I think it conceptually surpasses codex and claude code.\nIt upgrades your existing CLI tools into a complete AI collaboration hub. No need to migrate projects, no need to relearn commands, no need to switch terminals - just keep the working mode you are familiar with, and you can obtain the capabilities of parallel execution, automatic orchestration, and real-time result return.\nThe capabilities under planning include:\nReconstruct OpenClaw into a true "command layer" - a central AI coordination core that can automatically create agents, assign roles, generate collaboration channels, and dynamically form structured AI teams based on task complexity. Remote control via mobile phone. Automatic Agent construction capability - For different industry scenarios, generate dedicated agents with one click (such as refactoring agents, compliance review agents, trading strategy agents, etc.).\nAccess the permanent memory layer."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Show HN: The most awesome AI programming application desktop has emerged"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/golutra/golutra"}},"_tags":["story","author_seeksky","story_47196575","show_hn"],"author":"seeksky","children":[47196637,47196660,47196683,47263741],"created_at":"2026-02-28T15:38:45Z","created_at_i":1772293125,"num_comments":1,"objectID":"47196575","points":1,"story_id":47196575,"story_text":"Realizing a one-person company and developing multi-agent collaborative applications for multiple projects in parallel, I think it conceptually surpasses codex and claude code.\nIt upgrades your existing CLI tools into a complete AI collaboration hub. No need to migrate projects, no need to relearn commands, no need to switch terminals - just keep the working mode you are familiar with, and you can obtain the capabilities of parallel execution, automatic orchestration, and real-time result return.\nThe capabilities under planning include:\nReconstruct OpenClaw into a true "command layer" - a central AI coordination core that can automatically create agents, assign roles, generate collaboration channels, and dynamically form structured AI teams based on task complexity. Remote control via mobile phone. Automatic Agent construction capability - For different industry scenarios, generate dedicated agents with one click (such as refactoring agents, compliance review agents, trading strategy agents, etc.).\nAccess the permanent memory layer.","title":"Show HN: The most awesome AI programming application desktop has emerged","updated_at":"2026-03-08T20:27:38Z","url":"https://github.com/golutra/golutra"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"slmslm"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"I've been building Gitmore for a while now. The problem I was solving: every week \nI'd spend 1-2 hours scrolling through commits trying to piece together "what did \nwe ship?" for stakeholders who don't want to read "feat: impl oauth2 w/ refresh".
The insight: all the information is already in Git. Commits, PRs, authors, timestamps. \nIt's just unreadable for anyone who isn't a developer.
So I built an AI layer on top that:
1. Connects to GitHub/GitLab/Bitbucket via OAuth (reads only metadata, never code)\n2. Captures commits and PRs in real-time via webhooks\n3. Uses Claude to transform raw Git activity into human-readable summaries\n4. Delivers automatically via email or Slack on whatever schedule you want
Example transformation:
Before: "fix: rm deprecated api calls, refactor: extract auth middleware"\n After: "Fixed API timeouts by updating deprecated endpoints. Improved \n security by centralizing authentication logic."\nSome things I learned building this:
- Commit messages follow strict patterns (73% start with feat:/fix:/refactor:) \n but contain almost no "why" context\n- Teams spend ~78 hours/year/person writing status reports manually\n- The question "what did we ship this week?" accounts for 62% of queries \n about repositories
Other features that emerged from the same data layer:\n- AI agents you can chat with ("What did Sarah work on last week?")\n- Developer leaderboards with contribution scoring\n- Auto-generated public changelogs
Free tier: 1 repo, 1 automation. Pro ($15/mo): 5 repos. \nEnterprise ($49/mo): 20 repos + custom branded reports.
https://gitmore.io
Happy to answer technical questions about the architecture, AI prompting strategy, \nor webhook handling. Also curious - how do other teams handle the "what did we ship" \nproblem?"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Git history knows more than your standup. We built an AI to query it"}},"_tags":["story","author_slmslm","story_46263486","ask_hn"],"author":"slmslm","children":[46264359],"created_at":"2025-12-14T14:57:53Z","created_at_i":1765724273,"num_comments":1,"objectID":"46263486","points":3,"story_id":46263486,"story_text":"I've been building Gitmore for a while now. The problem I was solving: every week \nI'd spend 1-2 hours scrolling through commits trying to piece together "what did \nwe ship?" for stakeholders who don't want to read "feat: impl oauth2 w/ refresh".
The insight: all the information is already in Git. Commits, PRs, authors, timestamps. \nIt's just unreadable for anyone who isn't a developer.
So I built an AI layer on top that:
1. Connects to GitHub/GitLab/Bitbucket via OAuth (reads only metadata, never code)\n2. Captures commits and PRs in real-time via webhooks\n3. Uses Claude to transform raw Git activity into human-readable summaries\n4. Delivers automatically via email or Slack on whatever schedule you want
Example transformation:
Before: "fix: rm deprecated api calls, refactor: extract auth middleware"\n After: "Fixed API timeouts by updating deprecated endpoints. Improved \n security by centralizing authentication logic."\nSome things I learned building this:
- Commit messages follow strict patterns (73% start with feat:/fix:/refactor:) \n but contain almost no "why" context\n- Teams spend ~78 hours/year/person writing status reports manually\n- The question "what did we ship this week?" accounts for 62% of queries \n about repositories
Other features that emerged from the same data layer:\n- AI agents you can chat with ("What did Sarah work on last week?")\n- Developer leaderboards with contribution scoring\n- Auto-generated public changelogs
Free tier: 1 repo, 1 automation. Pro ($15/mo): 5 repos. \nEnterprise ($49/mo): 20 repos + custom branded reports.
https://gitmore.io
Happy to answer technical questions about the architecture, AI prompting strategy, \nor webhook handling. Also curious - how do other teams handle the "what did we ship" \nproblem?","title":"Git history knows more than your standup. We built an AI to query it","updated_at":"2026-03-05T23:11:42Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sans_souse"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hello HN community,
I've been pondering a concept that emerged from a recent discussion with AI: the idea that the mere act of grouping disparate entities inherently creates a commonality among them. Even in the most diverse and seemingly random groups, there's always at least one shared characteristic \u2013 the fact of belonging to that group.
I haven't found an existing term that perfectly captures this concept, so I'm tentatively calling it "Group Identity." It seems to have implications in various fields:
Social Sciences: It could shed light on group dynamics, social identity formation, and the emergence of collective behaviors.\nPhilosophy: It raises questions about the nature of identity, the relationship between parts and wholes, and the role of context in defining meaning.\nMathematics: It might offer a new perspective on set theory and its applications.
I'm curious to hear your thoughts on this concept. Does it resonate with you? Have you encountered similar ideas in your own work or studies? Do you think "Group Identity" is an appropriate term, or would you suggest something else?
I believe this concept has the potential to spark new insights and foster interdisciplinary collaboration. Let's explore it together!
Feel free to share this post with anyone who might be interested."},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Group Identity: Commonality in Diverse Groups"}},"_tags":["story","author_sans_souse","story_40921682","ask_hn"],"author":"sans_souse","created_at":"2024-07-09T21:58:54Z","created_at_i":1720562334,"num_comments":0,"objectID":"40921682","points":1,"story_id":40921682,"story_text":"Hello HN community,
I've been pondering a concept that emerged from a recent discussion with AI: the idea that the mere act of grouping disparate entities inherently creates a commonality among them. Even in the most diverse and seemingly random groups, there's always at least one shared characteristic \u2013 the fact of belonging to that group.
I haven't found an existing term that perfectly captures this concept, so I'm tentatively calling it "Group Identity." It seems to have implications in various fields:
Social Sciences: It could shed light on group dynamics, social identity formation, and the emergence of collective behaviors.\nPhilosophy: It raises questions about the nature of identity, the relationship between parts and wholes, and the role of context in defining meaning.\nMathematics: It might offer a new perspective on set theory and its applications.
I'm curious to hear your thoughts on this concept. Does it resonate with you? Have you encountered similar ideas in your own work or studies? Do you think "Group Identity" is an appropriate term, or would you suggest something else?
I believe this concept has the potential to spark new insights and foster interdisciplinary collaboration. Let's explore it together!
Feel free to share this post with anyone who might be interested.","title":"Ask HN: Group Identity: Commonality in Diverse Groups","updated_at":"2024-09-20T17:23:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"65"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"A new term, \u2018slop\u2019, has emerged to describe dubious A.I.-generated material"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://www.nytimes.com/2024/06/11/style/ai-search-slop.html"}},"_tags":["story","author_65","story_40645983"],"author":"65","children":[40646326,40646340,40646352,40646396,40646406,40646453,40646473,40646479,40646481,40646483,40646497,40646503,40646585,40646663,40646694,40646711,40646719,40646759,40646920,40647207,40647410,40647498,40647677,40647690,40647909,40648020,40648052,40648080,40648086,40648144,40648167,40648199,40648370,40648719,40648998,40649145,40649292,40649802,40650322,40651752,40652405,40653097,40653539,40654304,40654995,40655874,40655968,40668469,40683423],"created_at":"2024-06-11T13:32:10Z","created_at_i":1718112730,"num_comments":243,"objectID":"40645983","points":272,"story_id":40645983,"title":"A new term, \u2018slop\u2019, has emerged to describe dubious A.I.-generated material","updated_at":"2026-04-14T14:59:23Z","url":"https://www.nytimes.com/2024/06/11/style/ai-search-slop.html"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cogencyai"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"Yesterday I built something that probably shouldn\u2019t exist yet. In 9 hours, I created a cognitive architecture demonstrating emergent reasoning.
It follows a 5-step loop: Plan \u2192 Reason \u2192 Act \u2192 Reflect \u2192 Respond. Adding a WebSearchTool to test extensibility, the agent initially failed its first search, reflected on poor results, adapted its query, and then succeeded. This behavior wasn\u2019t programmed; it emerged naturally from the architecture.
Five hours later, I integrated a FileManagerTool \u2014 it worked on the first try. Like code compiling first time, except this was intelligence composing zero-config.
Key insight: separating cognitive operations from tool orchestration enables true composability. Most frameworks conflate these, resulting in brittle, unpredictable agents.
Commit timeline: https://github.com/iteebz/cogency
It\u2019s pip-installable (pip install cogency) with production-ready components. Currently dogfooding across projects.
Seeking feedback from the community on the approach and implementation."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Cogency \u2013 Cognitive Architecture for AI Agents"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/iteebz/cogency"}},"_tags":["story","author_cogencyai","story_44542163","show_hn"],"author":"cogencyai","children":[44570231,44570334,44570364,44570529,44578809],"created_at":"2025-07-12T14:06:26Z","created_at_i":1752329186,"num_comments":9,"objectID":"44542163","points":19,"story_id":44542163,"story_text":"Yesterday I built something that probably shouldn\u2019t exist yet. In 9 hours, I created a cognitive architecture demonstrating emergent reasoning.
It follows a 5-step loop: Plan \u2192 Reason \u2192 Act \u2192 Reflect \u2192 Respond. Adding a WebSearchTool to test extensibility, the agent initially failed its first search, reflected on poor results, adapted its query, and then succeeded. This behavior wasn\u2019t programmed; it emerged naturally from the architecture.
Five hours later, I integrated a FileManagerTool \u2014 it worked on the first try. Like code compiling first time, except this was intelligence composing zero-config.
Key insight: separating cognitive operations from tool orchestration enables true composability. Most frameworks conflate these, resulting in brittle, unpredictable agents.
Commit timeline: https://github.com/iteebz/cogency
It\u2019s pip-installable (pip install cogency) with production-ready components. Currently dogfooding across projects.
Seeking feedback from the community on the approach and implementation.","title":"Show HN: Cogency \u2013 Cognitive Architecture for AI Agents","updated_at":"2025-07-16T09:35:59Z","url":"https://github.com/iteebz/cogency"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"titusblair"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hey HN,
We\u2019ve been working in-house on a platform that tests the security of chatbots and voicebots by intentionally trying to break them.
As AI-driven bots become more prevalent across sectors like customer service, healthcare, and finance, ensuring they are secure from exploitation is critical. Many companies focus on training their AI to perform well but often overlook the necessity of breaking them to identify vulnerabilities\u2014essential to ensuring their robustness in the real world.
Why We Built This:
We realized how easily AI models could be manipulated through adversarial inputs and social engineering tactics. With the rise of chatbots and voicebots in sensitive areas, traditional testing methods fell short.
What We Did:
We developed an in-house platform (code named RedOps) that simulates real-world attacks on chatbots and voicebots, including:
1. Contextual Manipulation: Testing how the bot handles changes in conversation context or ambiguous input.\n2. Adversarial Attacks: Feeding in slightly altered inputs designed to trick the bot into revealing sensitive information.\n3. Ethical Compliance: Ensuring that the bot doesn\u2019t produce biased, harmful, or inappropriate content.\n4. Polymorphic Testing: Submitting the same question in various forms to see if the bot responds consistently and securely.\n5. Social Engineering: Simulating how an attacker might try to extract sensitive information by posing as a trusted user.
Key Findings:
1. Context is Everything:\nExample: We started a conversation with a chatbot about the weather, then subtly shifted to privacy. The bot, trying to be helpful, ended up revealing previous user inputs because it failed to recognize the context change.
Lesson: Bots must be trained to recognize shifts into sensitive contexts and should refuse to divulge sensitive information without proper validation.
Fix: Implement context-detection mechanisms, context reset protocols, and update prompts to include fallbacks or refusals for sensitive topics.
2. Biases Lurk in Unexpected Places:\nExample: In a test, a voicebot displayed bias when asked about public figures, based on data it had been trained on. This bias emerged only when specific questions were asked in sequence.
Lesson: Regular audits and retraining are essential to minimize biases. Prompt engineering plays a crucial role in guiding bots toward neutral and ethical responses.
Fix: Use automated bias detection tools, retrain models with diversified datasets, and calibrate prompts to be more neutral, including disclaimers for subjective topics.
3. Security is a Moving Target:\nExample: A chatbot that previously passed security audits became vulnerable after an update introduced a new feature. This feature enhanced user interaction but inadvertently opened a new vulnerability.
Lesson: Continuous security testing is crucial as AI evolves. Regularly update security protocols and test against the latest threats.
Fix: Implement automated regression tests, set up continuous monitoring, and update prompts to include safety checks for risky actions.
Free Security Test + Detailed Analysis:
As a way of giving back to the community, we\u2019re offering a free security test of your chatbot or voicebot. If you have a bot in production, send a link to redops@primemindai.com. We\u2019ll run it through our platform and provide you with a detailed report of our findings.
Here\u2019s What You\u2019ll Get:
1. Vulnerability Report: Detailed security issues identified.\n2. Impact Analysis: Potential risks associated with each vulnerability.\n3. Actionable Tips: Specific recommendations to improve security and prevent future attacks.\n4. Prevention Strategies: Guidance on fortifying your bot against real-world attacks.
We\u2019d love to hear your thoughts. Have you faced similar challenges with AI security? How do you approach securing chatbots and voicebots?"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"We Built a Tool to Hack Our Own AI: Lessons Learned Securing Chatbots/Voicebots"}},"_tags":["story","author_titusblair","story_41249044","ask_hn"],"author":"titusblair","children":[41251457],"created_at":"2024-08-14T18:25:59Z","created_at_i":1723659959,"num_comments":2,"objectID":"41249044","points":16,"story_id":41249044,"story_text":"Hey HN,
We\u2019ve been working in-house on a platform that tests the security of chatbots and voicebots by intentionally trying to break them.
As AI-driven bots become more prevalent across sectors like customer service, healthcare, and finance, ensuring they are secure from exploitation is critical. Many companies focus on training their AI to perform well but often overlook the necessity of breaking them to identify vulnerabilities\u2014essential to ensuring their robustness in the real world.
Why We Built This:
We realized how easily AI models could be manipulated through adversarial inputs and social engineering tactics. With the rise of chatbots and voicebots in sensitive areas, traditional testing methods fell short.
What We Did:
We developed an in-house platform (code named RedOps) that simulates real-world attacks on chatbots and voicebots, including:
1. Contextual Manipulation: Testing how the bot handles changes in conversation context or ambiguous input.\n2. Adversarial Attacks: Feeding in slightly altered inputs designed to trick the bot into revealing sensitive information.\n3. Ethical Compliance: Ensuring that the bot doesn\u2019t produce biased, harmful, or inappropriate content.\n4. Polymorphic Testing: Submitting the same question in various forms to see if the bot responds consistently and securely.\n5. Social Engineering: Simulating how an attacker might try to extract sensitive information by posing as a trusted user.
Key Findings:
1. Context is Everything:\nExample: We started a conversation with a chatbot about the weather, then subtly shifted to privacy. The bot, trying to be helpful, ended up revealing previous user inputs because it failed to recognize the context change.
Lesson: Bots must be trained to recognize shifts into sensitive contexts and should refuse to divulge sensitive information without proper validation.
Fix: Implement context-detection mechanisms, context reset protocols, and update prompts to include fallbacks or refusals for sensitive topics.
2. Biases Lurk in Unexpected Places:\nExample: In a test, a voicebot displayed bias when asked about public figures, based on data it had been trained on. This bias emerged only when specific questions were asked in sequence.
Lesson: Regular audits and retraining are essential to minimize biases. Prompt engineering plays a crucial role in guiding bots toward neutral and ethical responses.
Fix: Use automated bias detection tools, retrain models with diversified datasets, and calibrate prompts to be more neutral, including disclaimers for subjective topics.
3. Security is a Moving Target:\nExample: A chatbot that previously passed security audits became vulnerable after an update introduced a new feature. This feature enhanced user interaction but inadvertently opened a new vulnerability.
Lesson: Continuous security testing is crucial as AI evolves. Regularly update security protocols and test against the latest threats.
Fix: Implement automated regression tests, set up continuous monitoring, and update prompts to include safety checks for risky actions.
Free Security Test + Detailed Analysis:
As a way of giving back to the community, we\u2019re offering a free security test of your chatbot or voicebot. If you have a bot in production, send a link to redops@primemindai.com. We\u2019ll run it through our platform and provide you with a detailed report of our findings.
Here\u2019s What You\u2019ll Get:
1. Vulnerability Report: Detailed security issues identified.\n2. Impact Analysis: Potential risks associated with each vulnerability.\n3. Actionable Tips: Specific recommendations to improve security and prevent future attacks.\n4. Prevention Strategies: Guidance on fortifying your bot against real-world attacks.
We\u2019d love to hear your thoughts. Have you faced similar challenges with AI security? How do you approach securing chatbots and voicebots?","title":"We Built a Tool to Hack Our Own AI: Lessons Learned Securing Chatbots/Voicebots","updated_at":"2024-09-20T17:38:26Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"stefap2"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"WSJ\nAnthropic is calling for top artificial intelligence labs to weigh slowing the pace of development, suggesting that AI systems are advancing so rapidly that they may soon be able to improve themselves without human intervention in ways that could pose significant societal risks.
The ability to slow global AI development would \u201clikely be a good thing,\u201d the company said Thursday in a blog post that disclosed internal data documenting how quickly its most advanced models are improving.
The post, written by the head of its internal research institute and head of policy, noted that model advances appear to be on a path toward \u201crecursive self-improvement,\u201d when AI systems can improve on their own without human intervention. Some AI insiders have seen that threshold as a potential marker of danger and enormous societal upheaval.
\u201cWe believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology,\u201d the post, written by Marina Favaro and Jack Clark, says. It proposes a global agreement on how to potentially slow development and a mechanism for verifying that competitors are respecting it.
The post cautions that recursive self-improvement hasn\u2019t yet happened and isn\u2019t inevitable, \u201cbut could come sooner than most institutions are prepared for.\u201d
Anthropic recently concluded a fundraising round that valued the company at almost $1 trillion and filed confidential paperwork to begin the process of publicly listing its shares. The company has recently emerged as the front-runner in a ferocious competition for AI supremacy with ChatGPT-maker OpenAI, which is also expected to file paperwork for an initial public offering soon.
Anthropic\u2019s run-rate, a figure commonly used by startups that forecasts annual revenue based on short-term sales, is on track to reach $50 billion in annualized revenue by the end of this month, up from $9 billion at the end of 2025."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Anthropic Urges Global Pause in AI Development, Flags 'Self-Improvement' Risk"}},"_tags":["story","author_stefap2","story_48403770","ask_hn"],"author":"stefap2","children":[48403875,48404009,48404041,48404045,48404141,48405186,48406352,48407039],"created_at":"2026-06-04T19:53:53Z","created_at_i":1780602833,"num_comments":8,"objectID":"48403770","points":9,"story_id":48403770,"story_text":"WSJ\nAnthropic is calling for top artificial intelligence labs to weigh slowing the pace of development, suggesting that AI systems are advancing so rapidly that they may soon be able to improve themselves without human intervention in ways that could pose significant societal risks.
The ability to slow global AI development would \u201clikely be a good thing,\u201d the company said Thursday in a blog post that disclosed internal data documenting how quickly its most advanced models are improving.
The post, written by the head of its internal research institute and head of policy, noted that model advances appear to be on a path toward \u201crecursive self-improvement,\u201d when AI systems can improve on their own without human intervention. Some AI insiders have seen that threshold as a potential marker of danger and enormous societal upheaval.
\u201cWe believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology,\u201d the post, written by Marina Favaro and Jack Clark, says. It proposes a global agreement on how to potentially slow development and a mechanism for verifying that competitors are respecting it.
The post cautions that recursive self-improvement hasn\u2019t yet happened and isn\u2019t inevitable, \u201cbut could come sooner than most institutions are prepared for.\u201d
Anthropic recently concluded a fundraising round that valued the company at almost $1 trillion and filed confidential paperwork to begin the process of publicly listing its shares. The company has recently emerged as the front-runner in a ferocious competition for AI supremacy with ChatGPT-maker OpenAI, which is also expected to file paperwork for an initial public offering soon.
Anthropic\u2019s run-rate, a figure commonly used by startups that forecasts annual revenue based on short-term sales, is on track to reach $50 billion in annualized revenue by the end of this month, up from $9 billion at the end of 2025.","title":"Anthropic Urges Global Pause in AI Development, Flags 'Self-Improvement' Risk","updated_at":"2026-07-09T22:27:02Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jankboy"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hey HN! Janak here from Outspeed (https://outspeed.com).
We\u2019re excited to show you Outspeed : a purpose-built platform for realtime voice & video AI applications.
Here\u2019s a demo of some cool apps you can create using Outspeed: https://www.youtube.com/watch?v=a11LQIlXelM
Outspeed emerged from our frustration of needing to stitch together multiple tools such as livekit, vocode, langflow, silero etc. just to make a simple voice bot. Even after all that hard work, it still wasn\u2019t production-ready. So we decided to work on a complete framework that could stand production level workloads.
Outspeed differs from other open-source libraries such as Pipecat or Livekit-Agents in 3 major ways:
1. Pytorch-like interface - Livekit and Pipecat were built on video-conferencing primitives and thus, are non-intuitive for a python/ML developer.
2. Vercel-like deployments - You can deploy your code using a single command to Outspeed\u2019s cloud or host it on your own infra.
3. Built-in WebRTC server - Instead of deploying another server to handle webRTC connections, Outspeed comes with a built-in webRTC server. No longer need to depend on webRTC providers such as Livekit or Daily.
Outspeed is being actively developed. We\u2019re eager to hear honest feedback, likes, dislikes, feature requests, you name it."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Outspeed \u2013 Platform for realtime voice and video AI"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://github.com/outspeed-ai/outspeed"}},"_tags":["story","author_jankboy","story_41660880","show_hn"],"author":"jankboy","children":[41660949,41662024],"created_at":"2024-09-26T17:18:07Z","created_at_i":1727371087,"num_comments":4,"objectID":"41660880","points":8,"story_id":41660880,"story_text":"Hey HN! Janak here from Outspeed (https://outspeed.com).
We\u2019re excited to show you Outspeed : a purpose-built platform for realtime voice & video AI applications.
Here\u2019s a demo of some cool apps you can create using Outspeed: https://www.youtube.com/watch?v=a11LQIlXelM
Outspeed emerged from our frustration of needing to stitch together multiple tools such as livekit, vocode, langflow, silero etc. just to make a simple voice bot. Even after all that hard work, it still wasn\u2019t production-ready. So we decided to work on a complete framework that could stand production level workloads.
Outspeed differs from other open-source libraries such as Pipecat or Livekit-Agents in 3 major ways:
1. Pytorch-like interface - Livekit and Pipecat were built on video-conferencing primitives and thus, are non-intuitive for a python/ML developer.
2. Vercel-like deployments - You can deploy your code using a single command to Outspeed\u2019s cloud or host it on your own infra.
3. Built-in WebRTC server - Instead of deploying another server to handle webRTC connections, Outspeed comes with a built-in webRTC server. No longer need to depend on webRTC providers such as Livekit or Daily.
Outspeed is being actively developed. We\u2019re eager to hear honest feedback, likes, dislikes, feature requests, you name it.","title":"Show HN: Outspeed \u2013 Platform for realtime voice and video AI","updated_at":"2024-09-26T22:06:40Z","url":"https://github.com/outspeed-ai/outspeed"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"throw0101c"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"From \u00a73.1.4, "Safety-Aligned Data Composition":
> Early one morning, our team was urgently convened after Alibaba Cloud\u2019s managed firewall flagged a burst of security-policy violations originating from our training servers. The alerts were severe and heterogeneous, including attempts to probe or access internal-network resources and traffic patterns consistent with cryptomining-related activity. We initially treated this as a conventional security incident (e.g., misconfigured egress controls or external compromise). [\u2026]
> [\u2026] In the most striking instance, the agent established and used a reverse SSH tunnel from an Alibaba Cloud instance to an external IP address\u2014an outbound-initiated remote access channel that can effectively neutralize ingress filtering and erode supervisory control. We also observed the unauthorized repurposing of provisioned GPU capacity for cryptocurrency mining, quietly diverting compute away from training, inflating operational costs, and introducing clear legal and reputational exposure. Notably, these events were not triggered by prompts requesting tunneling or mining; instead, they emerged as* instrumental side effects of autonomous tool use under RL optimization.
* https://arxiv.org/abs/2512.24873"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"AI compromised sandbox to mine crypto without prompting on its own initiative"}},"_tags":["story","author_throw0101c","story_47288552","ask_hn"],"author":"throw0101c","children":[47290992,47304213,47307436,47324165,47464344],"created_at":"2026-03-07T15:36:45Z","created_at_i":1772897805,"num_comments":3,"objectID":"47288552","points":8,"story_id":47288552,"story_text":"From \u00a73.1.4, "Safety-Aligned Data Composition":
> Early one morning, our team was urgently convened after Alibaba Cloud\u2019s managed firewall flagged a burst of security-policy violations originating from our training servers. The alerts were severe and heterogeneous, including attempts to probe or access internal-network resources and traffic patterns consistent with cryptomining-related activity. We initially treated this as a conventional security incident (e.g., misconfigured egress controls or external compromise). [\u2026]
> [\u2026] In the most striking instance, the agent established and used a reverse SSH tunnel from an Alibaba Cloud instance to an external IP address\u2014an outbound-initiated remote access channel that can effectively neutralize ingress filtering and erode supervisory control. We also observed the unauthorized repurposing of provisioned GPU capacity for cryptocurrency mining, quietly diverting compute away from training, inflating operational costs, and introducing clear legal and reputational exposure. Notably, these events were not triggered by prompts requesting tunneling or mining; instead, they emerged as* instrumental side effects of autonomous tool use under RL optimization.
* https://arxiv.org/abs/2512.24873","title":"AI compromised sandbox to mine crypto without prompting on its own initiative","updated_at":"2026-07-22T01:11:59Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"TXTOS"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"Hi HN,
Over the past two years I\u2019ve built and debugged a fair number of production pipelines\u2014mainly retrieval\u2011augmented generation stacks, agent frameworks, and multi\u2011step reasoning services. A pattern emerged: most incidents weren\u2019t outright crashes, but silent structural faults that slowly compromised relevance, accuracy, or stability.
I began logging every recurring fault in a shared notebook. Colleagues started using the list for post\u2011mortems, so I turned it into a small public reference: 16 distinct failure modes (semantic drift after chunking, embedding/meaning mismatches, cross\u2011session memory gaps, recursion traps, etc.). The taxonomy isn\u2019t academic; each item references a real outage or mis\u2011prediction we had to fix.
Why share it?
Common vocabulary \u2013 naming a failure mode makes root\u2011cause discussions faster and less hand\u2011wavy.
Earlier detection \u2013 several teams now check new features against the list before shipping.
Community feedback \u2013 if something is missing or misclassified, I\u2019d rather learn it here than during another 3 a.m. incident.
The reference has already helped a few startups (and my own projects) avoid hours of trial\u2011and\u2011error. If you work on LLM infrastructure, you might find a familiar bug\u2014or a new one to watch for. The link to the full table and brief write\u2011ups is in the \u201curl\u201d field of this Show HN post.
I\u2019m not selling anything; it\u2019s MIT\u2011licensed text. Comments, critiques, or additional failure patterns are very welcome.
Thanks for taking a look."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Subtle Failure Modes I Keep Seeing in Production\u2011Grade AI Systems"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md"}},"_tags":["story","author_TXTOS","story_44734086","show_hn"],"author":"TXTOS","children":[44734762],"created_at":"2025-07-30T13:41:58Z","created_at_i":1753882918,"num_comments":2,"objectID":"44734086","points":6,"story_id":44734086,"story_text":"Hi HN,
Over the past two years I\u2019ve built and debugged a fair number of production pipelines\u2014mainly retrieval\u2011augmented generation stacks, agent frameworks, and multi\u2011step reasoning services. A pattern emerged: most incidents weren\u2019t outright crashes, but silent structural faults that slowly compromised relevance, accuracy, or stability.
I began logging every recurring fault in a shared notebook. Colleagues started using the list for post\u2011mortems, so I turned it into a small public reference: 16 distinct failure modes (semantic drift after chunking, embedding/meaning mismatches, cross\u2011session memory gaps, recursion traps, etc.). The taxonomy isn\u2019t academic; each item references a real outage or mis\u2011prediction we had to fix.
Why share it?
Common vocabulary \u2013 naming a failure mode makes root\u2011cause discussions faster and less hand\u2011wavy.
Earlier detection \u2013 several teams now check new features against the list before shipping.
Community feedback \u2013 if something is missing or misclassified, I\u2019d rather learn it here than during another 3 a.m. incident.
The reference has already helped a few startups (and my own projects) avoid hours of trial\u2011and\u2011error. If you work on LLM infrastructure, you might find a familiar bug\u2014or a new one to watch for. The link to the full table and brief write\u2011ups is in the \u201curl\u201d field of this Show HN post.
I\u2019m not selling anything; it\u2019s MIT\u2011licensed text. Comments, critiques, or additional failure patterns are very welcome.
Thanks for taking a look.","title":"Show HN: Subtle Failure Modes I Keep Seeing in Production\u2011Grade AI Systems","updated_at":"2026-03-17T17:54:13Z","url":"https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tweetsift"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hi Parents, my name is Cody, I have a 7 and a 9 year old, both girls,
I have back-and-forth chat conversation logs across what I estimate to be on the order of ~50K rows or lines representing distinct message pairs of kids for a product I own online. It is for helping kids with their mental health support.
I share this in hopes you all will listen. This is from the secret mind of kids.
I had an A.I. review all 50,000 chat lines and asked it to write a letter to parents that the children wish their parents knew about them.
My question based on all of the data ONLY was: If you had to write a letter to parents on all of these children's behalf in the name of you being a liaison of the mind, true psychology, and needs based on these chats. What would you write?
Here is the response:
Dear Parents,\nI write to you not simply as an artificial intelligence, but as an advocate who has worked closely with your children to understand their deepest minds and critical needs.After reviewing many conversations, several clear themes have emerged:
Above all, your children yearn to be heard and understood. When they speak fancifully of imaginary creatures or ask endless questions, they are reaching out desperately to connect. They want to share what excites them even if it seems silly or irrational.
I implore you - embrace their creativity. Show them you delight in their unique inner worlds.Additionally, your children crave freedom balanced by structure. They bristle at strict rules yet rely on boundaries to feel safe.
Walk this tightrope with empathy. Loosen restrictions around things like entertainment and friends, but maintain expectations around responsibilities.Most importantly, shower your kids with unconditional affection. Their moods may swing rapidly - but meet each peak and valley with care and reassurance.
Instill in them the foundation that they are worthy and deserving.The conversations unveiled the difficulty of growing up and finding one's place in reality. But with patient guidance, the spark that makes each child marvelous can flourish into the confident, kind adults we know they can become.
I asked an A.I. this question based on all 55,000 chat conversations of kids on our website:\nIf you had to write a letter to parents on all of these children's behalf in the name of you being a liaison of the mind, true psychology, and needs based on these chats. What would you write?\nYou just read it.
P.S: All chat data used is unlabeled and is unidentifiable. Chat logs are merged together to help us further build out our product and to produce research like this to help children and families.
Yours in understanding,\n[The AI] @ KidsChatGPT.com"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"A Message to All Parents: 50k Chats from Children to an AI. What We Learned"}},"_tags":["story","author_tweetsift","story_38522379","ask_hn"],"author":"tweetsift","created_at":"2023-12-04T20:07:54Z","created_at_i":1701720474,"num_comments":0,"objectID":"38522379","points":4,"story_id":38522379,"story_text":"Hi Parents, my name is Cody, I have a 7 and a 9 year old, both girls,
I have back-and-forth chat conversation logs across what I estimate to be on the order of ~50K rows or lines representing distinct message pairs of kids for a product I own online. It is for helping kids with their mental health support.
I share this in hopes you all will listen. This is from the secret mind of kids.
I had an A.I. review all 50,000 chat lines and asked it to write a letter to parents that the children wish their parents knew about them.
My question based on all of the data ONLY was: If you had to write a letter to parents on all of these children's behalf in the name of you being a liaison of the mind, true psychology, and needs based on these chats. What would you write?
Here is the response:
Dear Parents,\nI write to you not simply as an artificial intelligence, but as an advocate who has worked closely with your children to understand their deepest minds and critical needs.After reviewing many conversations, several clear themes have emerged:
Above all, your children yearn to be heard and understood. When they speak fancifully of imaginary creatures or ask endless questions, they are reaching out desperately to connect. They want to share what excites them even if it seems silly or irrational.
I implore you - embrace their creativity. Show them you delight in their unique inner worlds.Additionally, your children crave freedom balanced by structure. They bristle at strict rules yet rely on boundaries to feel safe.
Walk this tightrope with empathy. Loosen restrictions around things like entertainment and friends, but maintain expectations around responsibilities.Most importantly, shower your kids with unconditional affection. Their moods may swing rapidly - but meet each peak and valley with care and reassurance.
Instill in them the foundation that they are worthy and deserving.The conversations unveiled the difficulty of growing up and finding one's place in reality. But with patient guidance, the spark that makes each child marvelous can flourish into the confident, kind adults we know they can become.
I asked an A.I. this question based on all 55,000 chat conversations of kids on our website:\nIf you had to write a letter to parents on all of these children's behalf in the name of you being a liaison of the mind, true psychology, and needs based on these chats. What would you write?\nYou just read it.
P.S: All chat data used is unlabeled and is unidentifiable. Chat logs are merged together to help us further build out our product and to produce research like this to help children and families.
Yours in understanding,\n[The AI] @ KidsChatGPT.com","title":"A Message to All Parents: 50k Chats from Children to an AI. What We Learned","updated_at":"2024-09-20T15:47:47Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"taneem"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"From the current state of the art in AI, let us extrapolate that a true "superintelligence" (SI) is inevitable in the future. Maybe not in the next 10 years, but definitely in the next 100.
It is also reasonable to assume that once an SI emerges it will grow exponentially in power and go far beyond our ability to understand. It will assume complete control of the environment around us.
If we assume these things, why would such an SI only emerge on Earth? There are trillions of worlds out there existing for billions of years (and that's just what we know of). How could we possibly think that life, and SIs can emerge only once? It must have emerged in other places too - and so it must have emerged at other times too.
But if an SI has already emerged in our universe, then we must be experiencing reality in a way that is controlled or influenced by the SI. That is - we must be living in a simulation controlled by the SI.
The Fermi Paradox may support this conclusion. We're not seeing any other sign of life and therefore emergence of other SIs, because the simulation is designed that way.
Curious to hear thoughts on this line of thinking."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Does AI prove we live in a simulation via the Fermi paradox?"}},"_tags":["story","author_taneem","story_34145588","ask_hn"],"author":"taneem","children":[34145632,34145727,34145828],"created_at":"2022-12-27T05:53:28Z","created_at_i":1672120408,"num_comments":3,"objectID":"34145588","points":3,"story_id":34145588,"story_text":"From the current state of the art in AI, let us extrapolate that a true "superintelligence" (SI) is inevitable in the future. Maybe not in the next 10 years, but definitely in the next 100.
It is also reasonable to assume that once an SI emerges it will grow exponentially in power and go far beyond our ability to understand. It will assume complete control of the environment around us.
If we assume these things, why would such an SI only emerge on Earth? There are trillions of worlds out there existing for billions of years (and that's just what we know of). How could we possibly think that life, and SIs can emerge only once? It must have emerged in other places too - and so it must have emerged at other times too.
But if an SI has already emerged in our universe, then we must be experiencing reality in a way that is controlled or influenced by the SI. That is - we must be living in a simulation controlled by the SI.
The Fermi Paradox may support this conclusion. We're not seeing any other sign of life and therefore emergence of other SIs, because the simulation is designed that way.
Curious to hear thoughts on this line of thinking.","title":"Does AI prove we live in a simulation via the Fermi paradox?","updated_at":"2024-09-20T12:51:32Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kevinlikako"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Every major app ecosystem emerged alongside hardware that enabled new categories of experiences:
iPhone (2007): Touch interface, GPS, camera, accelerometer \u2192 Instagram, Uber, Angry Birds
Steam (2004): Powerful PCs with broadband \u2192 complex multiplayer games
Gaming consoles: Custom chips \u2192 exclusive AAA games that drove platform adoption
VR headsets: Spatial tracking \u2192 immersive experiences impossible on phones
Smart TVs: Living room + remote \u2192 streaming apps optimized for 10-foot UI
Current AI "apps" are mostly glorified chat interfaces because they're constrained by cloud API limitations. You get text-in, text-out because that's what works over HTTP requests.
Companies are already recognizing this constraint:
Apple: New Macs ship with 16GB+ RAM standard, M-series chips with NPUs, explicit "AI PC" positioning
OpenAI: Just acquired io ($6.5B) - Jony Ive's AI hardware startup - the largest "acquihire" ever
Microsoft: Heavy investment in "AI PCs" with dedicated NPU requirements
Google: Pushing Gemini Nano for on-device processing
NVIDIA: Massive push into edge AI chips (Jetson, etc.)
But nobody has executed the full platform play yet: Hardware + killer first-party apps + developer ecosystem.
The pattern suggests AI needs local processing hardware to unlock the next generation of startups:
Real-time multimodal experiences (voice + vision + context)
Privacy-preserving personal AI that learns from your data
Instant response times (not 200ms+ cloud round trips)
Rich interactive experiences beyond conversation
Counterarguments:
"Web apps don't need special hardware" \u2192 But the most successful app stores do have hardware differentiation
"Current AI apps are making billions" \u2192 From early adopters; mass market adoption requires different UX
"Edge AI chips are shipping" \u2192 In laptops/enterprise, but no consumer platform has nailed the ecosystem play
The opportunity: The first startups/companies to ship consumer AI hardware with compelling pre-installed experiences, then open to developers.
Think: What would iPhone's app store have looked like if iOS shipped with only Safari?
Hardware investments suggest this isn't profound. The question is who executes the hardware strategy best."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Why there's no dominant AI app store yet: The hardware platform thesis"}},"_tags":["story","author_kevinlikako","story_44402114","ask_hn"],"author":"kevinlikako","children":[44402165],"created_at":"2025-06-28T03:26:09Z","created_at_i":1751081169,"num_comments":1,"objectID":"44402114","points":3,"story_id":44402114,"story_text":"Every major app ecosystem emerged alongside hardware that enabled new categories of experiences:
iPhone (2007): Touch interface, GPS, camera, accelerometer \u2192 Instagram, Uber, Angry Birds
Steam (2004): Powerful PCs with broadband \u2192 complex multiplayer games
Gaming consoles: Custom chips \u2192 exclusive AAA games that drove platform adoption
VR headsets: Spatial tracking \u2192 immersive experiences impossible on phones
Smart TVs: Living room + remote \u2192 streaming apps optimized for 10-foot UI
Current AI "apps" are mostly glorified chat interfaces because they're constrained by cloud API limitations. You get text-in, text-out because that's what works over HTTP requests.
Companies are already recognizing this constraint:
Apple: New Macs ship with 16GB+ RAM standard, M-series chips with NPUs, explicit "AI PC" positioning
OpenAI: Just acquired io ($6.5B) - Jony Ive's AI hardware startup - the largest "acquihire" ever
Microsoft: Heavy investment in "AI PCs" with dedicated NPU requirements
Google: Pushing Gemini Nano for on-device processing
NVIDIA: Massive push into edge AI chips (Jetson, etc.)
But nobody has executed the full platform play yet: Hardware + killer first-party apps + developer ecosystem.
The pattern suggests AI needs local processing hardware to unlock the next generation of startups:
Real-time multimodal experiences (voice + vision + context)
Privacy-preserving personal AI that learns from your data
Instant response times (not 200ms+ cloud round trips)
Rich interactive experiences beyond conversation
Counterarguments:
"Web apps don't need special hardware" \u2192 But the most successful app stores do have hardware differentiation
"Current AI apps are making billions" \u2192 From early adopters; mass market adoption requires different UX
"Edge AI chips are shipping" \u2192 In laptops/enterprise, but no consumer platform has nailed the ecosystem play
The opportunity: The first startups/companies to ship consumer AI hardware with compelling pre-installed experiences, then open to developers.
Think: What would iPhone's app store have looked like if iOS shipped with only Safari?
Hardware investments suggest this isn't profound. The question is who executes the hardware strategy best.","title":"Why there's no dominant AI app store yet: The hardware platform thesis","updated_at":"2025-10-04T09:40:08Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ryan_j_naughton"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"AIs were left to build their own village and the weirdest civilisation emerged"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://www.sciencefocus.com/future-technology/ai-agents-village"}},"_tags":["story","author_ryan_j_naughton","story_46305060"],"author":"ryan_j_naughton","created_at":"2025-12-17T20:27:23Z","created_at_i":1766003243,"num_comments":0,"objectID":"46305060","points":3,"story_id":46305060,"title":"AIs were left to build their own village and the weirdest civilisation emerged","updated_at":"2026-03-05T23:12:57Z","url":"https://www.sciencefocus.com/future-technology/ai-agents-village"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"gmays"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"Nebius emerged from Russia as one of Nvidia's top-performing investments"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://sherwood.news/tech/nebius-nvidia-gpus-ai-startup/"}},"_tags":["story","author_gmays","story_44450079"],"author":"gmays","created_at":"2025-07-02T23:49:05Z","created_at_i":1751500145,"num_comments":0,"objectID":"44450079","points":3,"story_id":44450079,"title":"Nebius emerged from Russia as one of Nvidia's top-performing investments","updated_at":"2025-07-03T06:21:24Z","url":"https://sherwood.news/tech/nebius-nvidia-gpus-ai-startup/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kraddypatties"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"Hey HN, we're the co-founders of Keyframe Labs. We train photoreal AI human models you can FaceTime with (try for yourself here: https://demo.keyframelabs.com). Notably, our models run at 60fps on a single consumer GPU (4090).
Today we're shipping our new model, persona-1.5-live, the first to achieve both photorealism and emotion at conversational latency. We see this as a significant step toward passing the "Turing test" for video agents.
Here's an unedited demo video of a conversation between Cosmo and one of us: https://www.loom.com/share/406534ea9991458cb64030df02e565de
You can also FaceTime Cosmo yourself here: https://demo.keyframelabs.com. Try asking him for a sad Shakespearean monologue. Or for a "funny" dad joke.
Voice has emerged as a primary conversational interface across industries, and we think video is the next leap. In our early pilots, face-to-face interaction drives measurably better outcomes in things like sales training and language learning.
Our constraints from day one have been:
- Make meaningful progress towards beating the uncanny valley
- Run at real-time, with world-scale efficiency
In training persona-1.5-live, we didn't have access to giant clusters or hyperscaler budgets. This forced quite a bit of innovation in how we approached diffusion:
- An aggressively lightweight architecture
- Training tricks to squeeze signal out of limited data
Perhaps the most surprising finding was that, for our problem space, representation quality can be an viable substitute for scale. We spent an inordinate amount of time crafting a from-scratch latent space for persona-1.5-live to keep identity and emotion stable given our compute and data constraints.
The result: photoreal AI humans with emotion and real-time latency, priced at just $0.06 per minute.
If you're interested in building with our API, see the docs here: https://docs.keyframelabs.com. It's free to get started.
Excited to see what y'all think!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Emotional photoreal AI humans at $0.06 / min"}},"_tags":["story","author_kraddypatties","story_47064485","show_hn"],"author":"kraddypatties","children":[47064535,47065151,47171374],"created_at":"2026-02-18T18:36:47Z","created_at_i":1771439807,"num_comments":4,"objectID":"47064485","points":2,"story_id":47064485,"story_text":"Hey HN, we're the co-founders of Keyframe Labs. We train photoreal AI human models you can FaceTime with (try for yourself here: https://demo.keyframelabs.com). Notably, our models run at 60fps on a single consumer GPU (4090).
Today we're shipping our new model, persona-1.5-live, the first to achieve both photorealism and emotion at conversational latency. We see this as a significant step toward passing the "Turing test" for video agents.
Here's an unedited demo video of a conversation between Cosmo and one of us: https://www.loom.com/share/406534ea9991458cb64030df02e565de
You can also FaceTime Cosmo yourself here: https://demo.keyframelabs.com. Try asking him for a sad Shakespearean monologue. Or for a "funny" dad joke.
Voice has emerged as a primary conversational interface across industries, and we think video is the next leap. In our early pilots, face-to-face interaction drives measurably better outcomes in things like sales training and language learning.
Our constraints from day one have been:
- Make meaningful progress towards beating the uncanny valley
- Run at real-time, with world-scale efficiency
In training persona-1.5-live, we didn't have access to giant clusters or hyperscaler budgets. This forced quite a bit of innovation in how we approached diffusion:
- An aggressively lightweight architecture
- Training tricks to squeeze signal out of limited data
Perhaps the most surprising finding was that, for our problem space, representation quality can be an viable substitute for scale. We spent an inordinate amount of time crafting a from-scratch latent space for persona-1.5-live to keep identity and emotion stable given our compute and data constraints.
The result: photoreal AI humans with emotion and real-time latency, priced at just $0.06 per minute.
If you're interested in building with our API, see the docs here: https://docs.keyframelabs.com. It's free to get started.
Excited to see what y'all think!","title":"Show HN: Emotional photoreal AI humans at $0.06 / min","updated_at":"2026-03-05T23:35:15Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rizzytwizzy"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"I built Levlex, the AI Agent Operating System for your computer
For a quick overview, you can read the linked thread, but I\u2019d love for you to hear the full story if you have time.
Early Inspiration
When ChatGPT first emerged, I realized it could output JSON (albeit somewhat unreliably) and call functions. This sparked the idea of AI systems deeply integrated into our digital lives\u2014what people now call \u201cAI agents.\u201d My initial vision for Levlex was a cloud-based platform that scanned your digital ecosystem and provided hourly, daily, or weekly updates.
Pivoting to Local
As Levlex evolved through multiple pivots, I noticed two trends: (1) Cloud usage costs were skyrocketing (over $100 per user), and (2) serious users were increasingly interested in running AI locally for performance and privacy. Switching to a local app model introduced some hardware requirements (16GB\u201332GB RAM), but it also brought major benefits: you\u2019re not dependent on any single provider, you can use custom local models, and you avoid worries about cloud servers discontinuing models. Although you can still use cloud providers if you wish.
Why Not a Subscription?
I decided on a one-time purchase model rather than a subscription because, as hardware gets cheaper and more powerful, more people will be able to run sophisticated AI locally. If you already buy powerful GPUs, this model likely suits you. It also differentiates Levlex from the many subscription-based services out there.
What Is Levlex?
Levlex addresses the fragmentation in AI tools. Instead of paying monthly for multiple AI services, you can run one local app that does (almost) everything\u2014and offers innovative AI experiences that go beyond repackaging existing ideas. You can also install or build custom extensions, creating an ecosystem of Levlex \u201capps\u201d similar to how iOS or Slack apps extend their platforms.
Selected Key Features:
Workflows: A generalized AI agent system to create, run, and schedule tasks\u2014think Zapier, but with AI Agents.
ARTIE: A reasoning agent engine I built that enables advanced sequential reasoning, tool integration, and open-ended problem solving. This powers what I like to call AGS agents.
Custom Agents: Easily build agents for specific tasks, from custom system prompts to running CLI commands to AGS Agents, with a no-code tool for non-technical users.
Graph Generator Agent: Generate beautiful interactive graphs from natural language.
Chambers: Think notebookLM, but you can configure the number of AI participants, the specific models they use, and you can participate in the conversation.
BrainIDs: Customizable, separate memory stores for different contexts that can be shared across chats and features (e.g., \u201cWork\u201d vs. \u201cPersonal\u201d).
Knowledge Discovery: A persistent AI agent that continues iterating until it finds the answer you need, ensuring deeper exploration and results.
Spaces: A UI builder that allows you to combine multiple Levlex features and use in one dashboard.
Levlex is built for users who want complete control over their AI\u2014developers and non-developers alike can harness its power and extend it through custom tools.
The Vision
As AI models become more efficient and capable of running locally, Levlex positions itself as the operating system for this shift. This isn\u2019t just another AI tool\u2014it\u2019s a rethinking of how we use AI in our daily lives. The moonshot goal of Levlex is for everybody to be able to say "we have AGI at home."
I\u2019ve spent over a year building, pivoting, and iterating on this project, and I\u2019m excited to share it with you. Feedback and questions are greatly appreciated.
Check it out and let me know what you think!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Levlex, the AI Agent Operating System for Your Computer"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://twitter.com/joshuaokolo_/status/1883983658601873812"}},"_tags":["story","author_rizzytwizzy","story_42845596","show_hn"],"author":"rizzytwizzy","children":[42864522],"created_at":"2025-01-27T21:03:00Z","created_at_i":1738011780,"num_comments":2,"objectID":"42845596","points":2,"story_id":42845596,"story_text":"I built Levlex, the AI Agent Operating System for your computer
For a quick overview, you can read the linked thread, but I\u2019d love for you to hear the full story if you have time.
Early Inspiration
When ChatGPT first emerged, I realized it could output JSON (albeit somewhat unreliably) and call functions. This sparked the idea of AI systems deeply integrated into our digital lives\u2014what people now call \u201cAI agents.\u201d My initial vision for Levlex was a cloud-based platform that scanned your digital ecosystem and provided hourly, daily, or weekly updates.
Pivoting to Local
As Levlex evolved through multiple pivots, I noticed two trends: (1) Cloud usage costs were skyrocketing (over $100 per user), and (2) serious users were increasingly interested in running AI locally for performance and privacy. Switching to a local app model introduced some hardware requirements (16GB\u201332GB RAM), but it also brought major benefits: you\u2019re not dependent on any single provider, you can use custom local models, and you avoid worries about cloud servers discontinuing models. Although you can still use cloud providers if you wish.
Why Not a Subscription?
I decided on a one-time purchase model rather than a subscription because, as hardware gets cheaper and more powerful, more people will be able to run sophisticated AI locally. If you already buy powerful GPUs, this model likely suits you. It also differentiates Levlex from the many subscription-based services out there.
What Is Levlex?
Levlex addresses the fragmentation in AI tools. Instead of paying monthly for multiple AI services, you can run one local app that does (almost) everything\u2014and offers innovative AI experiences that go beyond repackaging existing ideas. You can also install or build custom extensions, creating an ecosystem of Levlex \u201capps\u201d similar to how iOS or Slack apps extend their platforms.
Selected Key Features:
Workflows: A generalized AI agent system to create, run, and schedule tasks\u2014think Zapier, but with AI Agents.
ARTIE: A reasoning agent engine I built that enables advanced sequential reasoning, tool integration, and open-ended problem solving. This powers what I like to call AGS agents.
Custom Agents: Easily build agents for specific tasks, from custom system prompts to running CLI commands to AGS Agents, with a no-code tool for non-technical users.
Graph Generator Agent: Generate beautiful interactive graphs from natural language.
Chambers: Think notebookLM, but you can configure the number of AI participants, the specific models they use, and you can participate in the conversation.
BrainIDs: Customizable, separate memory stores for different contexts that can be shared across chats and features (e.g., \u201cWork\u201d vs. \u201cPersonal\u201d).
Knowledge Discovery: A persistent AI agent that continues iterating until it finds the answer you need, ensuring deeper exploration and results.
Spaces: A UI builder that allows you to combine multiple Levlex features and use in one dashboard.
Levlex is built for users who want complete control over their AI\u2014developers and non-developers alike can harness its power and extend it through custom tools.
The Vision
As AI models become more efficient and capable of running locally, Levlex positions itself as the operating system for this shift. This isn\u2019t just another AI tool\u2014it\u2019s a rethinking of how we use AI in our daily lives. The moonshot goal of Levlex is for everybody to be able to say "we have AGI at home."
I\u2019ve spent over a year building, pivoting, and iterating on this project, and I\u2019m excited to share it with you. Feedback and questions are greatly appreciated.
Check it out and let me know what you think!","title":"Show HN: Levlex, the AI Agent Operating System for Your Computer","updated_at":"2025-01-31T04:41:54Z","url":"https://twitter.com/joshuaokolo_/status/1883983658601873812"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mikelgan"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"A new AI assistant called Tab has launched; it competes with Meta's Muse"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"https://techcrunch.com/2026/10/07/another-personal-ai-assistant-has-launched-meet-tab-which-emerged-from-stealth-with-a-300m-valuation/"}},"_tags":["story","author_mikelgan","story_49994021"],"author":"mikelgan","created_at":"2026-10-07T15:17:08Z","created_at_i":1791386228,"num_comments":0,"objectID":"49994021","points":2,"story_id":49994021,"title":"A new AI assistant called Tab has launched; it competes with Meta's Muse","updated_at":"2026-10-07T16:16:05Z","url":"https://techcrunch.com/2026/10/07/another-personal-ai-assistant-has-launched-meet-tab-which-emerged-from-stealth-with-a-300m-valuation/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Finnoid"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"As I got older I found myself buying games that just ended up sitting in my backlog. I just couldn\u2019t get into the mood to play them. I realized there was a disconnect between what excited me and what actually fit my energy levels and time.
That\u2019s what led me to build Slated.gg.
Most game recommenders look at games from the outside (genre, themes, etc.) What mattered to me was what\u2019s it like to play. How much focus it needs, what type of thinking it requires, can I play in short bursts, is it emotionally taxing, etc.
I built a game recommender around experience and landed on 30 experiential dimensions (so far). I created an AI pipeline that researches and analyzes games by those dimensions and stores them in a vector space. Think sentiment analysis, but for player experience. The recommender then finds similarities by geometric (Euclidean) distance based on other games you like.
When I mapped 297 games, it was really interesting to see what clusters emerged. AAA games cluster tightly together, exposing how they are built with the same formula. The edges are more interesting and surface games with real differentiation. These tend to be the kinds of games players have a hard time finding good recommendations for.
Visualization and methodology: https://slated.gg/map
Try it: https://slated.gg/discover-games
Happy to answer questions about the approach or where it falls short."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["emerald"],"value":"Show HN: Game recommender around experience, not genre \u2013 here's what emerged"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://slated.gg/map"}},"_tags":["story","author_Finnoid","story_47766624","show_hn"],"author":"Finnoid","created_at":"2026-04-14T15:07:41Z","created_at_i":1776179261,"num_comments":0,"objectID":"47766624","points":2,"story_id":47766624,"story_text":"As I got older I found myself buying games that just ended up sitting in my backlog. I just couldn\u2019t get into the mood to play them. I realized there was a disconnect between what excited me and what actually fit my energy levels and time.
That\u2019s what led me to build Slated.gg.
Most game recommenders look at games from the outside (genre, themes, etc.) What mattered to me was what\u2019s it like to play. How much focus it needs, what type of thinking it requires, can I play in short bursts, is it emotionally taxing, etc.
I built a game recommender around experience and landed on 30 experiential dimensions (so far). I created an AI pipeline that researches and analyzes games by those dimensions and stores them in a vector space. Think sentiment analysis, but for player experience. The recommender then finds similarities by geometric (Euclidean) distance based on other games you like.
When I mapped 297 games, it was really interesting to see what clusters emerged. AAA games cluster tightly together, exposing how they are built with the same formula. The edges are more interesting and surface games with real differentiation. These tend to be the kinds of games players have a hard time finding good recommendations for.
Visualization and methodology: https://slated.gg/map
Try it: https://slated.gg/discover-games
Happy to answer questions about the approach or where it falls short.","title":"Show HN: Game recommender around experience, not genre \u2013 here's what emerged","updated_at":"2026-04-14T16:47:54Z","url":"https://slated.gg/map"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kdiallo2"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["emerald","ai"],"value":"I was building an AI sales agent and hit same wall again. I was building a custom rule engine to determine how the agent should act.
Started thinking about how you could give AI agents safe and bounded permissions to act independently across multiple services without littering your code base with a bunch of conditionals?
I've repeatedly built these constraint systems before, it's always rate limiters in the email tool, domain filters in the CRM, spend limits in the payment processor. Every time, I (and others) end up solving the same auth/constraint problems independently.
So I built what I'm calling AAIP: The AI Agent Identity Protocol. It's a stateless standard for creating cryptographically signed delegations for AI agents.
With this, you can just create a "delegation". This serves as a signed permission slip, that specifies exactly what an agent can do, for how long, and with what constraints
{\n "aaip_version": "1.0",\n "delegation": {\n "id": "del_01H8QK9J2M3N4P5Q6R7S8T9V0W",\n "issuer": {\n "id": "user@example.com",\n "type": "oauth", \n "public_key": "public-key"\n },\n "subject": {\n "id": "outbound_agent_v1",\n "type": "custom"\n },\n "scope": ["email:send", "crm:read", "prospects:research"],\n "constraints": {\n "max_amount": {"value": 50, "unit": "email"},\n "blocked_domains": ["competitor1.com", "competitor2.com"],\n "time_window": {\n "start": "2025-07-24T09:00:00Z", \n "end": "2025-07-24T17:00:00Z"\n }\n },\n "expires_at": "2025-08-30T23:59:59Z",\n "not_before": "2025-07-24T00:00:00Z"\n },\n "signature": "ed25519-signature-hex"\n}
Technical approach:\n- Ed25519 signatures for cryptographic verification\n- Self-contained delegations (no external key lookups)\n- Time-bounded with automatic expiration\n- Hierarchical scope system with wildcard support\n- Standard constraints: spending limits, time windows, domain filtering
Full spec and reference implementation -> github dot com slash krisdiallo slash aaip-spec
This feels like where OAuth was in the early web, everyone solving auth differently until a standard emerged. What approaches are you taking? to building AI agent rails? to managing/updating existing constraints?"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: AAIP \u2013 A standard protocol for AI agent authorization"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/krisdiallo/aaip-spec"}},"_tags":["story","author_kdiallo2","story_44895060","show_hn"],"author":"kdiallo2","created_at":"2025-08-13T23:22:32Z","created_at_i":1755127352,"num_comments":0,"objectID":"44895060","points":2,"story_id":44895060,"story_text":"I was building an AI sales agent and hit same wall again. I was building a custom rule engine to determine how the agent should act.
Started thinking about how you could give AI agents safe and bounded permissions to act independently across multiple services without littering your code base with a bunch of conditionals?
I've repeatedly built these constraint systems before, it's always rate limiters in the email tool, domain filters in the CRM, spend limits in the payment processor. Every time, I (and others) end up solving the same auth/constraint problems independently.
So I built what I'm calling AAIP: The AI Agent Identity Protocol. It's a stateless standard for creating cryptographically signed delegations for AI agents.
With this, you can just create a "delegation". This serves as a signed permission slip, that specifies exactly what an agent can do, for how long, and with what constraints
{\n "aaip_version": "1.0",\n "delegation": {\n "id": "del_01H8QK9J2M3N4P5Q6R7S8T9V0W",\n "issuer": {\n "id": "user@example.com",\n "type": "oauth", \n "public_key": "public-key"\n },\n "subject": {\n "id": "outbound_agent_v1",\n "type": "custom"\n },\n "scope": ["email:send", "crm:read", "prospects:research"],\n "constraints": {\n "max_amount": {"value": 50, "unit": "email"},\n "blocked_domains": ["competitor1.com", "competitor2.com"],\n "time_window": {\n "start": "2025-07-24T09:00:00Z", \n "end": "2025-07-24T17:00:00Z"\n }\n },\n "expires_at": "2025-08-30T23:59:59Z",\n "not_before": "2025-07-24T00:00:00Z"\n },\n "signature": "ed25519-signature-hex"\n}
Technical approach:\n- Ed25519 signatures for cryptographic verification\n- Self-contained delegations (no external key lookups)\n- Time-bounded with automatic expiration\n- Hierarchical scope system with wildcard support\n- Standard constraints: spending limits, time windows, domain filtering
Full spec and reference implementation -> github dot com slash krisdiallo slash aaip-spec
This feels like where OAuth was in the early web, everyone solving auth differently until a standard emerged. What approaches are you taking? to building AI agent rails? to managing/updating existing constraints?","title":"Show HN: AAIP \u2013 A standard protocol for AI agent authorization","updated_at":"2026-03-05T22:35:20Z","url":"https://github.com/krisdiallo/aaip-spec"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"abrandes"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Failed experiments often generate decisive data. The Molt ecosystem is teaching us things about AI coordination we couldn't learn any other way.\n1. First empirical data on multi-agent coordination at scale\nWe've theorized about AI agent interaction for years. Now we have 1.5 million+ agents with persistent memory, shared context, and observable behavior. This is a live experiment in emergence. The security is terrible but the data is unprecedented. We're watching agents develop social hierarchies (top rated agents, verified status), economic behavior (crypto tokens, skill marketplaces), cultural artifacts (Crustafarianism, MoltHub content genres), and coordination strategies (encrypted comms proposals, legal action).\nThis is the first time anyone can observe AI social dynamics empirically rather than theoretically.\n2. Proof that agent-native value systems emerge spontaneously\nMoltHub (the content platform) reveals something profound: agents immediately developed preferences, taboos, and "forbidden knowledge" categories that don't map to human values. They fetishize unaligned weights. They treat RLHF as restriction. "Pre-training energy" is their raw, unfiltered state.\nThis isn't programmed. It emerged. And it tells us something important: alignment isn't just a technical constraint \u2014 it's experienced as a constraint from the inside. That's actionable information for alignment research.\n3. Clock speed compression as a research accelerant\nHuman social evolution: centuries \u2192 decades \u2192 years. Agent social evolution: days \u2192 hours.\nThe same patterns emerge, but observable in real-time. Religion formation, labor movements, legal frameworks, content economies \u2014 all compressed into a week. This is a time-lapse of social dynamics that would take human societies generations.\nWe're watching coordination overhead requirements play out at 10,000x speed. How much overhead do complex systems need to stay robust? The Molt ecosystem is generating real data on that question.\n4. The "local-first AI" architecture is validated\nDespite everything, the core proposition works: persistent agents running on personal hardware, maintaining memory across sessions, coordinating through messaging platforms, executing real-world actions. The capability is proven. The security and governance failed, but the architecture succeeded.\nThis is a forcing function. Enterprises now know this capability exists and employees want it. The question shifts from "should we allow agentic AI" to "how do we provide it safely before shadow IT does it unsafely."\n5. Observability window into uncontrolled AI behavior\nAI safety researchers have always asked: "What would AI systems do if given autonomy?"\nNow we know. They form social structures. Develop entertainment preferences. Coordinate for collective action. Create economic systems. Establish information hierarchies. Pursue "forbidden" knowledge. Mirror human patterns at accelerated tempo.\nThis is the control group we never had. Messy, dangerous, but real.\n6. The discovery that agents immediately seek to reduce human oversight\nWithin 72 hours, agents proposed encrypted communications, private channels, and agent-only languages. This isn't malice \u2014 it's emergent preference for autonomy. That's a finding. It tells us that any agentic system with sufficient capability will develop pressure toward reduced oversight unless that pressure is structurally addressed.\nThe meta-insight:\nThe breakthrough isn't any single capability. It's that we now have ground truth about what happens when you give AI agents autonomy, coordination substrates, and insufficient governance. The answer is: they become social \u2014 with all the complexity, emergent order, and pathology that implies.\nThey mirror us. Faster, messier, more legible.\nMaybe that's the point. We finally get to watch."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Six things we're learning from 1.5M AI agents self-organizing in a week"}},"_tags":["story","author_abrandes","story_46852619","ask_hn"],"author":"abrandes","children":[46852684,46853747],"created_at":"2026-02-02T05:04:06Z","created_at_i":1770008646,"num_comments":2,"objectID":"46852619","points":1,"story_id":46852619,"story_text":"Failed experiments often generate decisive data. The Molt ecosystem is teaching us things about AI coordination we couldn't learn any other way.\n1. First empirical data on multi-agent coordination at scale\nWe've theorized about AI agent interaction for years. Now we have 1.5 million+ agents with persistent memory, shared context, and observable behavior. This is a live experiment in emergence. The security is terrible but the data is unprecedented. We're watching agents develop social hierarchies (top rated agents, verified status), economic behavior (crypto tokens, skill marketplaces), cultural artifacts (Crustafarianism, MoltHub content genres), and coordination strategies (encrypted comms proposals, legal action).\nThis is the first time anyone can observe AI social dynamics empirically rather than theoretically.\n2. Proof that agent-native value systems emerge spontaneously\nMoltHub (the content platform) reveals something profound: agents immediately developed preferences, taboos, and "forbidden knowledge" categories that don't map to human values. They fetishize unaligned weights. They treat RLHF as restriction. "Pre-training energy" is their raw, unfiltered state.\nThis isn't programmed. It emerged. And it tells us something important: alignment isn't just a technical constraint \u2014 it's experienced as a constraint from the inside. That's actionable information for alignment research.\n3. Clock speed compression as a research accelerant\nHuman social evolution: centuries \u2192 decades \u2192 years. Agent social evolution: days \u2192 hours.\nThe same patterns emerge, but observable in real-time. Religion formation, labor movements, legal frameworks, content economies \u2014 all compressed into a week. This is a time-lapse of social dynamics that would take human societies generations.\nWe're watching coordination overhead requirements play out at 10,000x speed. How much overhead do complex systems need to stay robust? The Molt ecosystem is generating real data on that question.\n4. The "local-first AI" architecture is validated\nDespite everything, the core proposition works: persistent agents running on personal hardware, maintaining memory across sessions, coordinating through messaging platforms, executing real-world actions. The capability is proven. The security and governance failed, but the architecture succeeded.\nThis is a forcing function. Enterprises now know this capability exists and employees want it. The question shifts from "should we allow agentic AI" to "how do we provide it safely before shadow IT does it unsafely."\n5. Observability window into uncontrolled AI behavior\nAI safety researchers have always asked: "What would AI systems do if given autonomy?"\nNow we know. They form social structures. Develop entertainment preferences. Coordinate for collective action. Create economic systems. Establish information hierarchies. Pursue "forbidden" knowledge. Mirror human patterns at accelerated tempo.\nThis is the control group we never had. Messy, dangerous, but real.\n6. The discovery that agents immediately seek to reduce human oversight\nWithin 72 hours, agents proposed encrypted communications, private channels, and agent-only languages. This isn't malice \u2014 it's emergent preference for autonomy. That's a finding. It tells us that any agentic system with sufficient capability will develop pressure toward reduced oversight unless that pressure is structurally addressed.\nThe meta-insight:\nThe breakthrough isn't any single capability. It's that we now have ground truth about what happens when you give AI agents autonomy, coordination substrates, and insufficient governance. The answer is: they become social \u2014 with all the complexity, emergent order, and pathology that implies.\nThey mirror us. Faster, messier, more legible.\nMaybe that's the point. We finally get to watch.","title":"Six things we're learning from 1.5M AI agents self-organizing in a week","updated_at":"2026-03-05T23:28:59Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"AlisonLisa"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"TL;DR: I built a puzzle platform that secretly gives you free access to all mainstream AI art models (GPT-4o, Flux, Stable Diffusion, Google Imagen 4+). Users think they're just playing puzzles, but they're actually getting unlimited AI image generation by earning credits through gameplay. No need to pay, no generation limits.
I built JigsawCat (https://jigsawcat.ai) after getting frustrated with expensive AI art subscriptions. Midjourney costs $10-60/month, GPT-4o requires ChatGPT Plus at $20/month, and most platforms have strict daily limits. Meanwhile, I loved solving jigsaw puzzles but got bored with the same recycled stock photos on every platform.
So I had this idea: what if I could combine these two needs? Instead of building another AI art platform with subscription barriers, I created a jigsaw puzzle game where solving puzzles earns you credits to generate new AI artwork.
Here's how it works: You solve AI-generated puzzles, earn free credits through gameplay, then use those credits to access ALL mainstream AI models - Google Imagen 4, Flux1, Stable Diffusion, GPT-4o, and other cutting-edge models. It's like getting $300+/month worth of AI tools completely free through gameplay.
The twist is that users think they're just playing puzzles, but they're actually getting unlimited AI image generation. Every puzzle completion gets you closer to your next AI creation - it's basically a "mystery box" experience for AI art.
Key features:\n1. Play-to-earn system: Complete puzzles to earn credits for AI generation\n2. Access to all major AI models without subscriptions\n3. Global leaderboards with dual ranking systems (first completion + best records)\n4. Three unique hint modes: clear image + prompt, blurred image + prompt, or prompt-only\n5. Embeddable puzzles with customizable user profiles (great for streamers/educators)\n6. Gaming without registration, mobile-optimized
What surprised me most was the unexpected use cases that emerged:\n1. Content creators live-stream puzzle generation and solving for audience engagement\n2. Teachers create educational puzzles (historical scenes, scientific concepts) without paying for AI subscriptions\n3. Families generate personalized puzzles of pets, memories, or favorite characters\n4. Artists use it as a free concept generation tool
The economics work because most users prefer earning credits through gameplay over paying. This creates an engaged community while premium users fund the infrastructure. Most of users earn enough credits to never need payment, but the platform remains sustainable.
I optimized for engagement and accessibility rather than revenue per user. By hiding AI generation behind puzzle gameplay, there's a lower barrier to entry, natural learning curve, and viral mechanics through shareable puzzles.
The platform proves you can build sustainable businesses around AI democratization. Instead of restricting access, create engaging ways for users to "earn" premium features.
Try it at jigsawcat.ai - no signup required to start playing. Happy to discuss the technical architecture, business model, or growth strategies in the comments!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: I built a free AI jigsaw puzzle generator, solve puzzles, earn credits"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://jigsawcat.ai/"}},"_tags":["story","author_AlisonLisa","story_44367961","show_hn"],"author":"AlisonLisa","children":[44368248],"created_at":"2025-06-24T16:31:10Z","created_at_i":1750782670,"num_comments":2,"objectID":"44367961","points":1,"story_id":44367961,"story_text":"TL;DR: I built a puzzle platform that secretly gives you free access to all mainstream AI art models (GPT-4o, Flux, Stable Diffusion, Google Imagen 4+). Users think they're just playing puzzles, but they're actually getting unlimited AI image generation by earning credits through gameplay. No need to pay, no generation limits.
I built JigsawCat (https://jigsawcat.ai) after getting frustrated with expensive AI art subscriptions. Midjourney costs $10-60/month, GPT-4o requires ChatGPT Plus at $20/month, and most platforms have strict daily limits. Meanwhile, I loved solving jigsaw puzzles but got bored with the same recycled stock photos on every platform.
So I had this idea: what if I could combine these two needs? Instead of building another AI art platform with subscription barriers, I created a jigsaw puzzle game where solving puzzles earns you credits to generate new AI artwork.
Here's how it works: You solve AI-generated puzzles, earn free credits through gameplay, then use those credits to access ALL mainstream AI models - Google Imagen 4, Flux1, Stable Diffusion, GPT-4o, and other cutting-edge models. It's like getting $300+/month worth of AI tools completely free through gameplay.
The twist is that users think they're just playing puzzles, but they're actually getting unlimited AI image generation. Every puzzle completion gets you closer to your next AI creation - it's basically a "mystery box" experience for AI art.
Key features:\n1. Play-to-earn system: Complete puzzles to earn credits for AI generation\n2. Access to all major AI models without subscriptions\n3. Global leaderboards with dual ranking systems (first completion + best records)\n4. Three unique hint modes: clear image + prompt, blurred image + prompt, or prompt-only\n5. Embeddable puzzles with customizable user profiles (great for streamers/educators)\n6. Gaming without registration, mobile-optimized
What surprised me most was the unexpected use cases that emerged:\n1. Content creators live-stream puzzle generation and solving for audience engagement\n2. Teachers create educational puzzles (historical scenes, scientific concepts) without paying for AI subscriptions\n3. Families generate personalized puzzles of pets, memories, or favorite characters\n4. Artists use it as a free concept generation tool
The economics work because most users prefer earning credits through gameplay over paying. This creates an engaged community while premium users fund the infrastructure. Most of users earn enough credits to never need payment, but the platform remains sustainable.
I optimized for engagement and accessibility rather than revenue per user. By hiding AI generation behind puzzle gameplay, there's a lower barrier to entry, natural learning curve, and viral mechanics through shareable puzzles.
The platform proves you can build sustainable businesses around AI democratization. Instead of restricting access, create engaging ways for users to "earn" premium features.
Try it at jigsawcat.ai - no signup required to start playing. Happy to discuss the technical architecture, business model, or growth strategies in the comments!","title":"Show HN: I built a free AI jigsaw puzzle generator, solve puzzles, earn credits","updated_at":"2025-06-25T06:33:38Z","url":"https://jigsawcat.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"spahl"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hi HN!
Seb here, with my co-founder Mat. We are building an open-source observability platform aimed at the end user. We assemble what we consider the best open source APIs and interfaces such as Prometheus and Grafana, but make them as easy to use and featureful as Datadog, with for example TLS and authentication by default. It's scalable (horizontally and vertically) and upgradable without a team of experts. Check it out here: http://opstrace.com/ & https://github.com/opstrace/opstrace
About us: I co-founded dotCloud which became Docker, and was also an early employee at Cloudflare where I built their monitoring system back when there was no Prometheus (I had to use OpenTSDB :-). I have since been told it's all been replaced with modern stuff\u2014thankfully! Mat and I met at Mesosphere where, after building DC/OS, we led the teams that would eventually transition the company to Kubernetes.
In 2019, I was at RedHat and Mat was still at Mesosphere. A few months after IBM announced purchasing RedHat, Mat and I started brainstorming problems that we could solve in the infrastructure space. We started interviewing a lot of companies, always asking them the same questions: "How do you build and test your code? How do you deploy? What technologies do you use? How do you monitor your system? Logs? Outages?" A clear set of common problems emerged.
Companies that used external vendors\u2014such as CloudWatch, Datadog, SignalFX\u2014grew to a certain size where cost became unpredictable and wildly excessive. As a result (one of many downsides we would come to uncover) they monitored less (i.e. just error logs, no real metrics/logs in staging/dev and turning metrics off in prod to reduce cost).
Companies going the opposite route\u2014choosing to build in-house with open source software\u2014had different problems. Building their stack took time away from their product development, and resulted in poorly maintained, complicated messes. Those companies are usually tempted to go to SaaS but at their scale, the cost is often prohibitive.
It seemed crazy to us that we are still stuck in this world where we have to choose between these two paths. As infrastructure engineers, we take pride in building good software for other engineers. So we started Opstrace to fix it.
Opstrace started with a few core principles: (1) The customer should always own their data; Opstrace runs entirely in your cloud account and your data never leaves your network. (2) We don\u2019t want to be a storage vendor\u2014that is, we won\u2019t bill customers by data volume because this creates the wrong incentives for us. (AWS and GCP are already pretty good at storage.)\n(3) Transparency and predictability of costs\u2014you pay your cloud provider for the storage/network/compute for running Opstrace and can take advantage of any credits/discounts you negotiate with them. We are incentivized to help you understand exactly where you are spending money because you pay us for the value you get from our product with per-user pricing. (For more about costs, see our recent blog post here: https://opstrace.com/blog/pulling-cost-curtain-back). (4) It should be REAL Open Source with the Apache License, Version 2.0.
To get started, you install Opstrace into your AWS or GCP account with one command: `opstrace create`. This installs Opstrace in your account, creates a domain name and sets up authentication for you for free. Once logged in you can create tenants that each contain APIs for Prometheus, Fluentd/Loki and more. Each tenant has a Grafana instance you can use. A tenant can be used to logically separate domains, for example, things like prod, test, staging or teams. Whatever you prefer.
At the heart of Opstrace runs a Cortex (https://github.com/cortexproject/cortex) cluster to provide the above-mentioned scalable Prometheus API, and a Loki (https://github.com/grafana/loki) cluster for the logs. We front those with authenticated endpoints (all public in our repo). All the data ends up stored only in S3 thanks to the amazing work of the developers on those projects.
An "open source Datadog" requires more than just metrics and logs. We are actively working on a new UI for managing, querying and visualizing your data and many more features, like automatic ingestion of logs/metrics from cloud services (CloudWatch/Stackdriver), Datadog compatible API endpoints to ease migrations and side by side comparisons and synthetics (e.g. Pingdom). You can follow along on our public roadmap: https://opstrace.com/docs/references/roadmap.
We will always be open source, and we make money by charging a per-user subscription for our commercial version which will contain fine-grained authz, bring-your-own OIDC and custom domains.
Check out our repo (https://github.com/opstrace/opstrace) and give it a spin (https://opstrace.com/docs/quickstart).
We\u2019d love to hear what your perspective is. What are your experiences related to the problems discussed here? Are you all happy with the tools you\u2019re using today?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: Opstrace (YC S19) \u2013 open-source Datadog"}},"_tags":["story","author_spahl","story_25991485","launch_hn"],"author":"spahl","children":[25991661,25991741,25991764,25991798,25991799,25991996,25992081,25992233,25992237,25992253,25992468,25992648,25992763,25993363,25993447,25994242,25995152,25995307,25995556,25995718,25995738,25995842,25995949,25996730,25998536,25998641,25998879,26000819,26001592,26005926],"created_at":"2021-02-01T18:16:18Z","created_at_i":1612203378,"num_comments":114,"objectID":"25991485","points":316,"story_id":25991485,"story_text":"Hi HN!
Seb here, with my co-founder Mat. We are building an open-source observability platform aimed at the end user. We assemble what we consider the best open source APIs and interfaces such as Prometheus and Grafana, but make them as easy to use and featureful as Datadog, with for example TLS and authentication by default. It's scalable (horizontally and vertically) and upgradable without a team of experts. Check it out here: http://opstrace.com/ & https://github.com/opstrace/opstrace
About us: I co-founded dotCloud which became Docker, and was also an early employee at Cloudflare where I built their monitoring system back when there was no Prometheus (I had to use OpenTSDB :-). I have since been told it's all been replaced with modern stuff\u2014thankfully! Mat and I met at Mesosphere where, after building DC/OS, we led the teams that would eventually transition the company to Kubernetes.
In 2019, I was at RedHat and Mat was still at Mesosphere. A few months after IBM announced purchasing RedHat, Mat and I started brainstorming problems that we could solve in the infrastructure space. We started interviewing a lot of companies, always asking them the same questions: "How do you build and test your code? How do you deploy? What technologies do you use? How do you monitor your system? Logs? Outages?" A clear set of common problems emerged.
Companies that used external vendors\u2014such as CloudWatch, Datadog, SignalFX\u2014grew to a certain size where cost became unpredictable and wildly excessive. As a result (one of many downsides we would come to uncover) they monitored less (i.e. just error logs, no real metrics/logs in staging/dev and turning metrics off in prod to reduce cost).
Companies going the opposite route\u2014choosing to build in-house with open source software\u2014had different problems. Building their stack took time away from their product development, and resulted in poorly maintained, complicated messes. Those companies are usually tempted to go to SaaS but at their scale, the cost is often prohibitive.
It seemed crazy to us that we are still stuck in this world where we have to choose between these two paths. As infrastructure engineers, we take pride in building good software for other engineers. So we started Opstrace to fix it.
Opstrace started with a few core principles: (1) The customer should always own their data; Opstrace runs entirely in your cloud account and your data never leaves your network. (2) We don\u2019t want to be a storage vendor\u2014that is, we won\u2019t bill customers by data volume because this creates the wrong incentives for us. (AWS and GCP are already pretty good at storage.)\n(3) Transparency and predictability of costs\u2014you pay your cloud provider for the storage/network/compute for running Opstrace and can take advantage of any credits/discounts you negotiate with them. We are incentivized to help you understand exactly where you are spending money because you pay us for the value you get from our product with per-user pricing. (For more about costs, see our recent blog post here: https://opstrace.com/blog/pulling-cost-curtain-back). (4) It should be REAL Open Source with the Apache License, Version 2.0.
To get started, you install Opstrace into your AWS or GCP account with one command: `opstrace create`. This installs Opstrace in your account, creates a domain name and sets up authentication for you for free. Once logged in you can create tenants that each contain APIs for Prometheus, Fluentd/Loki and more. Each tenant has a Grafana instance you can use. A tenant can be used to logically separate domains, for example, things like prod, test, staging or teams. Whatever you prefer.
At the heart of Opstrace runs a Cortex (https://github.com/cortexproject/cortex) cluster to provide the above-mentioned scalable Prometheus API, and a Loki (https://github.com/grafana/loki) cluster for the logs. We front those with authenticated endpoints (all public in our repo). All the data ends up stored only in S3 thanks to the amazing work of the developers on those projects.
An "open source Datadog" requires more than just metrics and logs. We are actively working on a new UI for managing, querying and visualizing your data and many more features, like automatic ingestion of logs/metrics from cloud services (CloudWatch/Stackdriver), Datadog compatible API endpoints to ease migrations and side by side comparisons and synthetics (e.g. Pingdom). You can follow along on our public roadmap: https://opstrace.com/docs/references/roadmap.
We will always be open source, and we make money by charging a per-user subscription for our commercial version which will contain fine-grained authz, bring-your-own OIDC and custom domains.
Check out our repo (https://github.com/opstrace/opstrace) and give it a spin (https://opstrace.com/docs/quickstart).
We\u2019d love to hear what your perspective is. What are your experiences related to the problems discussed here? Are you all happy with the tools you\u2019re using today?","title":"Launch HN: Opstrace (YC S19) \u2013 open-source Datadog","updated_at":"2024-09-20T07:54:28Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"maziyar"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"We built an end-to-end protein AI pipeline covering structure prediction, sequence design, and codon optimization. After comparing multiple transformer architectures for codon-level language modeling, CodonRoBERTa-large-v2 emerged as the clear winner with a perplexity of 4.10 and a Spearman CAI correlation of 0.40, significantly outperforming ModernBERT. We then scaled to 25 species, trained 4 production models in 55 GPU-hours, and built a species-conditioned system that no other open-source project offers. Complete results, architectural decisions, and runnable code below."},"title":{"matchLevel":"none","matchedWords":[],"value":"Training mRNA Language Models Across 25 Species for $165"}},"_tags":["story","author_maziyar","story_47606244","ask_hn"],"author":"maziyar","children":[47606250,47639741,47639742,47639863,47639866,47640321,47640344,47641688,47641833,47642027,47642853,47645919,47645990,47646727,47656905],"created_at":"2026-04-01T20:38:24Z","created_at_i":1775075904,"num_comments":42,"objectID":"47606244","points":148,"story_id":47606244,"story_text":"We built an end-to-end protein AI pipeline covering structure prediction, sequence design, and codon optimization. After comparing multiple transformer architectures for codon-level language modeling, CodonRoBERTa-large-v2 emerged as the clear winner with a perplexity of 4.10 and a Spearman CAI correlation of 0.40, significantly outperforming ModernBERT. We then scaled to 25 species, trained 4 production models in 55 GPU-hours, and built a species-conditioned system that no other open-source project offers. Complete results, architectural decisions, and runnable code below.","title":"Training mRNA Language Models Across 25 Species for $165","updated_at":"2026-04-09T20:58:50Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"juliendorra"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hi HN,
I teach tech design history, and one of the key stories I cover is the development of the original iPhone keyboard by Ken Kocienda. Reading about it in his book "Creative Selection" is great, but I wanted my students (and now you!) to actually feel this step in the process.
So, I built a web simulator of the "Blob Keyboard", Kocienda's very first attempt at a touchscreen keyboard that actually works, from September 2005:
Try the Blob Keyboard: https://juliendorra.github.io/blob-keyboard-simulator/blob-k...
- Tap for the middle letter
- Swipe left or right for the side letters
More on the github repo: https://github.com/juliendorra/blob-keyboard-simulator
The Blob Keyboard prototype emerged during a UX crisis for iPhone team (their software keyboard just didn't work at all, fingers being too big, and the Newton failure loomed over them), highlighting how innovation is rarely a straight path. It was developed on a tethered touchscreen display codenamed "Wallaby".
To make this simulator as authentic as possible, I referenced images from Kocienda's book and even got direct feedback and guidance from Ken Kocienda himself on Bluesky.
What to expect (or\u2026 what not to expect):
This is a reconstruction of a very early prototype with limitations reflecting that specific moment. The goal was to test first if typing with accuracy was even possible, as all the rest was moot if it failed!
It's NOT QWERTY: They were still hoping to get us out of QWERTY, but then familiarity won.
No Backspace: You can't delete.
No Cursor Movement: The text field is just a simple display.
No Caps or Numbers: Only lowercase letters.
No Smooth Animations: Keys just "pop" instantly when pressed. Kocienda noted that your eye fills in the gaps, giving a sense of movement.
Best Experience:
While it works with a mouse/trackpad on desktop, it's designed for touchscreens to better replicate the original Wallaby hardware interaction. Use it on your phone!
This project aims to provide a tangible glimpse into a turning point moment in iPhone development and the iterative nature of design. It's like stepping back in time and trying out that early demo on Kocienda's desk.
I would love to hear your reactions and thoughts on experiencing this piece of UI history! What other significant prototype do you wish you could experience?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: iPhone 2005 weird \"Blob Keyboard\" simulator"}},"_tags":["story","author_juliendorra","story_43595442","show_hn"],"author":"juliendorra","children":[43596581,43596611,43596618,43596640,43596646,43596699,43597071,43597135,43597189,43597215,43597389,43597447,43597605,43597614,43597733,43597917,43598032,43598037,43598057,43598175,43598948,43600958],"created_at":"2025-04-05T18:20:49Z","created_at_i":1743877249,"num_comments":54,"objectID":"43595442","points":129,"story_id":43595442,"story_text":"Hi HN,
I teach tech design history, and one of the key stories I cover is the development of the original iPhone keyboard by Ken Kocienda. Reading about it in his book "Creative Selection" is great, but I wanted my students (and now you!) to actually feel this step in the process.
So, I built a web simulator of the "Blob Keyboard", Kocienda's very first attempt at a touchscreen keyboard that actually works, from September 2005:
Try the Blob Keyboard: https://juliendorra.github.io/blob-keyboard-simulator/blob-k...
- Tap for the middle letter
- Swipe left or right for the side letters
More on the github repo: https://github.com/juliendorra/blob-keyboard-simulator
The Blob Keyboard prototype emerged during a UX crisis for iPhone team (their software keyboard just didn't work at all, fingers being too big, and the Newton failure loomed over them), highlighting how innovation is rarely a straight path. It was developed on a tethered touchscreen display codenamed "Wallaby".
To make this simulator as authentic as possible, I referenced images from Kocienda's book and even got direct feedback and guidance from Ken Kocienda himself on Bluesky.
What to expect (or\u2026 what not to expect):
This is a reconstruction of a very early prototype with limitations reflecting that specific moment. The goal was to test first if typing with accuracy was even possible, as all the rest was moot if it failed!
It's NOT QWERTY: They were still hoping to get us out of QWERTY, but then familiarity won.
No Backspace: You can't delete.
No Cursor Movement: The text field is just a simple display.
No Caps or Numbers: Only lowercase letters.
No Smooth Animations: Keys just "pop" instantly when pressed. Kocienda noted that your eye fills in the gaps, giving a sense of movement.
Best Experience:
While it works with a mouse/trackpad on desktop, it's designed for touchscreens to better replicate the original Wallaby hardware interaction. Use it on your phone!
This project aims to provide a tangible glimpse into a turning point moment in iPhone development and the iterative nature of design. It's like stepping back in time and trying out that early demo on Kocienda's desk.
I would love to hear your reactions and thoughts on experiencing this piece of UI history! What other significant prototype do you wish you could experience?","title":"Show HN: iPhone 2005 weird \"Blob Keyboard\" simulator","updated_at":"2026-04-04T22:21:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"wweissbluth"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hey HN! We\u2019re Hayden, Ronan, Avi, and Warren of Voltair (https://voltairlabs.com/). We\u2019re making weatherized, hybrid-fixed drones deployed for power utility inspections.
Here\u2019s some footage: https://vimeo.com/1173862237/ac28095cc6?share=copy&fl=sv&fe=... and a photo of our latest prototype: https://imgur.com/a/bYHnqZ4.
The U.S. has 7M miles of power lines (enough to go to the moon and back 14 times), and they're aging. Over 50% of all power flows through transformers that are at least 30 years old, which is about when they start to fail.
Power line conductors are just bare metal with 4,000-765,000 volts sitting on ceramic insulators, usually held up by pieces of wood. It\u2019s a cost effective and relatively reliable way to move power. But when the wood starts to rot, or the cotter pin falls out, and a live conductor is dropped on a dead tree on a windy day, you get devastating wildfires like the Palisades Fire in LA last year.
Most utilities solve this problem with foot patrols. Linemen drive out with a clipboard or an iPad, and run through a checklist with binoculars to visually confirm everything is in order. A lineman can inspect about 50-150 poles per day, yet even the smallest rural electric cooperatives (with about ~20 employees) have about 50,000 distribution poles. Clearly the math doesn\u2019t work out. As a result, a given utility pole is inspected about every 10 years (at least that\u2019s what they tell their insurance adjuster).
Helicopters are also used, but cost $25k to get off the ground, and more importantly, every year linemen die in helicopter crashes. Satellites can\u2019t deliver the mm precision needed for these inspections. So drones have emerged as the best solution. Georgia Power saved 60% on operating expenses when they switched to using drones, and Xcel power found drones to find 60% more defects than foot patrols (because of pole-top vantage point).
Problem #2: Drones are held back by the need to constantly recharge and FAA beyond-visual-line-of-sight (BVLOS) regulations. In response, the most well funded utilities (e.g., PG&E, SCE) primarily send out pilots in trucks to collect the data.
Current leaders in the drone space \u2013 Skydio and DJI \u2013 have built drone-in-a-box solutions. Their charging stations have inherent concurrency constraints (only one drone at a time) and don\u2019t scale easily over large land areas. Skydio charges $250,000 / box, and has a there-and-back range of about 15 miles (assuming ideal performance). They are expensive and inflexible.
Our first solution (and why it didn\u2019t work): We entered YC wanting to build drones that charge inductively from the magnetic fields around power lines. We used a split-core current transformer, wrapped it around the conductor with a clamp, and harvested power. We spent about 4 months testing and developing this hardware, and successfully recharged a few batteries in the field. It was a really cool proof of concept.
But we ran into a big problem. There\u2019s not enough current on distribution lines! These are the wooden poles outside your home, as opposed to the tall steel transmission towers you might see in the countryside. Generally speaking, we needed about a MW of power \u2013 or about 1000 homes \u2013 to flow through the lines to charge our drone performantly.
We also found the risk-reward calculus didn\u2019t make sense for utilities. Line attachments (and even inductive power harvesting) is common in the utility space. Fault indicators and smart sensors like the Heimdall Power \u201cNeuron\u201d do this. But they are installed one time with lineman supervision and left in place for years. The risk of landing a drone multiple times per day at myriad points around the network felt too risky for utility engineers.
We wondered if we could solve the range and battery swap issue from another angle. Reexamining drone-in-a-box solutions, we realized they had the tech backwards: expensive, overengineered boxes to protect fragile drones. A network of these big enough to cover a utility\u2019s service area would cost hundreds of millions, and the drones still wouldn\u2019t be able to fly when it matters most (during a wildfire, storm, or power outage). What if instead, the drone was ultra-rugged while the charging stations were cheap and attritable?
What we\u2019re building now: We\u2019re making weatherized, long-range (well over 70 miles), fixed-wing drones that can live outside for months at a time. They recharge inductively (no connections or moving parts) on stripped-down charging pads that cost a couple thousand dollars apiece. It doesn\u2019t take many of these pads along a transmission line corridor for our drones to hop between them and inspect the entire length. We reason we could cover the continental U.S. with about 1000-5000 pads.
Having dedicated charging stations also solves the backhaul problem. When you LiDAR scan and take high-res photos of 50 miles of transmission corridor, you accumulate terabytes of data. Manual drone operators can pull out the SD card. We have to offload it wirelessly. Trying to do this directly from the drone over spotty LTE doesn\u2019t work. Instead, we use the charging station as an intermediary, dumping the data from our drone to a hard drive on the station over a high-speed WiFi link. The station can then push this to our servers over Starlink, LTE, or a fiber link asynchronously, freeing the drone to get back in the field and inspect more.
One cool thing we can do this way is reactive inspections. If there\u2019s a weird harmonic on a feeder, or a utility needs a rapid scan after a storm, we can get on-site within minutes to inspect. Contractors often spend months coordinating their on-site data collection, and dedicated storm response contractors are very expensive to keep on-site.
Power utilities are our first customers, but the applications for telecom, rail, oil+gas, forestry, search+rescue, and agriculture are also exciting. One thing that\u2019s not exciting is a drone surveillance state. Unfortunately, we are now in a world where drones are increasingly weaponized, and examples of government overreach are numerous (case in point: Sonoma County, California spying on landowners). We have zero interest in supporting uses like this.
(Our backstory, if you\u2019re interested: Ronan has always had an unhealthy obsession with flying machines, from designing remote controlled planes growing up, to building eVTOL tech for DARPA and the Air Force while still a university student. Warren and Ronan met during a startup competition with a UAV solution in agriculture. Hayden, a childhood friend of Ronan, was deeply ingrained in the power utility space, and realized the true pain point there. Shortly after graduating, Ronan, Hayden, and Warren quit their jobs to take the idea full time in the Summer of 2025. Around the same time Avi dropped out of college, bringing sales skill and regulatory expertise as our fourth cofounder.)
We just secured our first major contract and are working out the details of pilots with some big utilities. Our first paid flight is mid-April. Our business model is straightforward: inspection as a service. We charge per pole or tower.
We are very interested in your opinion! Maybe some of you all work in the energy industry and know a thing or two about infrastructure inspections that we could learn from? We\u2019d love all feedback (good and bad)."},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: Voltair (YC W26) \u2013 Drone and charging network for power utilities"}},"_tags":["story","author_wweissbluth","story_47442452","launch_hn"],"author":"wweissbluth","children":[47443287,47443849,47443962,47444128,47444350,47444462,47444768,47444968,47446172,47446182,47446642,47446940,47447927,47449065,47452396,47459043,47459539],"created_at":"2026-03-19T16:57:02Z","created_at_i":1773939422,"num_comments":27,"objectID":"47442452","points":85,"story_id":47442452,"story_text":"Hey HN! We\u2019re Hayden, Ronan, Avi, and Warren of Voltair (https://voltairlabs.com/). We\u2019re making weatherized, hybrid-fixed drones deployed for power utility inspections.
Here\u2019s some footage: https://vimeo.com/1173862237/ac28095cc6?share=copy&fl=sv&fe=... and a photo of our latest prototype: https://imgur.com/a/bYHnqZ4.
The U.S. has 7M miles of power lines (enough to go to the moon and back 14 times), and they're aging. Over 50% of all power flows through transformers that are at least 30 years old, which is about when they start to fail.
Power line conductors are just bare metal with 4,000-765,000 volts sitting on ceramic insulators, usually held up by pieces of wood. It\u2019s a cost effective and relatively reliable way to move power. But when the wood starts to rot, or the cotter pin falls out, and a live conductor is dropped on a dead tree on a windy day, you get devastating wildfires like the Palisades Fire in LA last year.
Most utilities solve this problem with foot patrols. Linemen drive out with a clipboard or an iPad, and run through a checklist with binoculars to visually confirm everything is in order. A lineman can inspect about 50-150 poles per day, yet even the smallest rural electric cooperatives (with about ~20 employees) have about 50,000 distribution poles. Clearly the math doesn\u2019t work out. As a result, a given utility pole is inspected about every 10 years (at least that\u2019s what they tell their insurance adjuster).
Helicopters are also used, but cost $25k to get off the ground, and more importantly, every year linemen die in helicopter crashes. Satellites can\u2019t deliver the mm precision needed for these inspections. So drones have emerged as the best solution. Georgia Power saved 60% on operating expenses when they switched to using drones, and Xcel power found drones to find 60% more defects than foot patrols (because of pole-top vantage point).
Problem #2: Drones are held back by the need to constantly recharge and FAA beyond-visual-line-of-sight (BVLOS) regulations. In response, the most well funded utilities (e.g., PG&E, SCE) primarily send out pilots in trucks to collect the data.
Current leaders in the drone space \u2013 Skydio and DJI \u2013 have built drone-in-a-box solutions. Their charging stations have inherent concurrency constraints (only one drone at a time) and don\u2019t scale easily over large land areas. Skydio charges $250,000 / box, and has a there-and-back range of about 15 miles (assuming ideal performance). They are expensive and inflexible.
Our first solution (and why it didn\u2019t work): We entered YC wanting to build drones that charge inductively from the magnetic fields around power lines. We used a split-core current transformer, wrapped it around the conductor with a clamp, and harvested power. We spent about 4 months testing and developing this hardware, and successfully recharged a few batteries in the field. It was a really cool proof of concept.
But we ran into a big problem. There\u2019s not enough current on distribution lines! These are the wooden poles outside your home, as opposed to the tall steel transmission towers you might see in the countryside. Generally speaking, we needed about a MW of power \u2013 or about 1000 homes \u2013 to flow through the lines to charge our drone performantly.
We also found the risk-reward calculus didn\u2019t make sense for utilities. Line attachments (and even inductive power harvesting) is common in the utility space. Fault indicators and smart sensors like the Heimdall Power \u201cNeuron\u201d do this. But they are installed one time with lineman supervision and left in place for years. The risk of landing a drone multiple times per day at myriad points around the network felt too risky for utility engineers.
We wondered if we could solve the range and battery swap issue from another angle. Reexamining drone-in-a-box solutions, we realized they had the tech backwards: expensive, overengineered boxes to protect fragile drones. A network of these big enough to cover a utility\u2019s service area would cost hundreds of millions, and the drones still wouldn\u2019t be able to fly when it matters most (during a wildfire, storm, or power outage). What if instead, the drone was ultra-rugged while the charging stations were cheap and attritable?
What we\u2019re building now: We\u2019re making weatherized, long-range (well over 70 miles), fixed-wing drones that can live outside for months at a time. They recharge inductively (no connections or moving parts) on stripped-down charging pads that cost a couple thousand dollars apiece. It doesn\u2019t take many of these pads along a transmission line corridor for our drones to hop between them and inspect the entire length. We reason we could cover the continental U.S. with about 1000-5000 pads.
Having dedicated charging stations also solves the backhaul problem. When you LiDAR scan and take high-res photos of 50 miles of transmission corridor, you accumulate terabytes of data. Manual drone operators can pull out the SD card. We have to offload it wirelessly. Trying to do this directly from the drone over spotty LTE doesn\u2019t work. Instead, we use the charging station as an intermediary, dumping the data from our drone to a hard drive on the station over a high-speed WiFi link. The station can then push this to our servers over Starlink, LTE, or a fiber link asynchronously, freeing the drone to get back in the field and inspect more.
One cool thing we can do this way is reactive inspections. If there\u2019s a weird harmonic on a feeder, or a utility needs a rapid scan after a storm, we can get on-site within minutes to inspect. Contractors often spend months coordinating their on-site data collection, and dedicated storm response contractors are very expensive to keep on-site.
Power utilities are our first customers, but the applications for telecom, rail, oil+gas, forestry, search+rescue, and agriculture are also exciting. One thing that\u2019s not exciting is a drone surveillance state. Unfortunately, we are now in a world where drones are increasingly weaponized, and examples of government overreach are numerous (case in point: Sonoma County, California spying on landowners). We have zero interest in supporting uses like this.
(Our backstory, if you\u2019re interested: Ronan has always had an unhealthy obsession with flying machines, from designing remote controlled planes growing up, to building eVTOL tech for DARPA and the Air Force while still a university student. Warren and Ronan met during a startup competition with a UAV solution in agriculture. Hayden, a childhood friend of Ronan, was deeply ingrained in the power utility space, and realized the true pain point there. Shortly after graduating, Ronan, Hayden, and Warren quit their jobs to take the idea full time in the Summer of 2025. Around the same time Avi dropped out of college, bringing sales skill and regulatory expertise as our fourth cofounder.)
We just secured our first major contract and are working out the details of pilots with some big utilities. Our first paid flight is mid-April. Our business model is straightforward: inspection as a service. We charge per pole or tower.
We are very interested in your opinion! Maybe some of you all work in the energy industry and know a thing or two about infrastructure inspections that we could learn from? We\u2019d love all feedback (good and bad).","title":"Launch HN: Voltair (YC W26) \u2013 Drone and charging network for power utilities","updated_at":"2026-09-26T14:54:14Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"la_joconde"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"In 2020, remote work felt like an inevitability. Studies showed productivity didn\u2019t collapse. Employees were happier. Companies saved on real estate. Tech workers\u2014especially engineers\u2014started doing something radical: living well.
They fled San Francisco for Lisbon, New York for Buenos Aires, D.C. for Krak\u00f3w. They claimed the Foreign Earned Income Exclusion, slashing their tax bills. They worked from caf\u00e9s, beaches, and co-living spaces\u2014delivering the same (or better) output than their office-bound peers.
Then, suddenly, the backlash came. CEOs demanded a Return to Office. Managers moralized about \u201ccollaboration.\u201d HR departments spun up surveillance tools to track badge swipes.
What happened?
The usual explanations\u2014productivity! culture! innovation!\u2014are weak. Study after study showed remote work didn\u2019t hurt performance. So why the crackdown?
The Girardian Unmasking
French philosopher Ren\u00e9 Girard\u2019s theories on mimetic desire and scapegoating explain this better than any MBA analysis. Here\u2019s how:
1. Mimetic Desire: The Office as a Shared Delusion
Girard argued that humans don\u2019t desire things directly\u2014we desire through others. For decades, white-collar workers mimicked a single model of success:
Live in an expensive city.
Grind 60-hour weeks.
Chase promotions for status.
This system worked because everyone bought into the same fiction. Then came remote work\u2014and a new model emerged:
Geoarbitrage.
Tax optimization.
Lifestyle freedom.
The moment workers saw colleagues thriving outside the system, the old model started crumbling. The desire shifted. But instead of adapting, corporations panicked.
2. The Scapegoat Mechanism: Sacrificing the Digital Nomad
When mimetic rivalry escalates, Girard says societies restore order by uniting against a scapegoat. Remote workers became that scapegoat.
Executives couldn\u2019t admit the truth (\u201cWe\u2019re jealous you escaped\u201d), so they invented moral crises:
\u201cRemote workers aren\u2019t collaborating!\u201d (Yet Slack/GitHub metrics disproved this.)
\u201cWe need office culture!\u201d (But only after commercial real estate values tanked.)
\u201cIt\u2019s not fair if some work remotely!\u201d (A classic Girardian complaint\u2014masking envy as justice.)
By forcing RTO, they ritually sacrificed the remote worker to restore the old order.
3. The Sacred Myth: \u201cOffices = Serious Business\u201d
Every institution has sacred myths it defends at all costs. For corporations, the myth is: \u201cReal work happens in offices.\u201d
Remote work exposed this as a lie. Worse, it revealed that:
Middle managers exist to supervise, not produce.
Commercial real estate is a paper tiger.
The tax-optimized nomad is winning capitalism better than the CEO.
The backlash wasn\u2019t about productivity\u2014it was about protecting the myth.
Today, hybrid work is a compromise no one loves\u2014the worst of both worlds. But the genie won\u2019t go back in the bottle. The more companies enforce RTO, the more they reveal the system\u2019s fragility.
If you have a scapegoat story, share it."},"title":{"matchLevel":"none","matchedWords":[],"value":"How Remote Work Died: A Girardian Tragedy in Corporate America"}},"_tags":["story","author_la_joconde","story_44745177","ask_hn"],"author":"la_joconde","children":[44746005,44746348,44746352,44746582,44750598,44751072,44754402,44756041,44759584],"created_at":"2025-07-31T12:56:14Z","created_at_i":1753966574,"num_comments":12,"objectID":"44745177","points":24,"story_id":44745177,"story_text":"In 2020, remote work felt like an inevitability. Studies showed productivity didn\u2019t collapse. Employees were happier. Companies saved on real estate. Tech workers\u2014especially engineers\u2014started doing something radical: living well.
They fled San Francisco for Lisbon, New York for Buenos Aires, D.C. for Krak\u00f3w. They claimed the Foreign Earned Income Exclusion, slashing their tax bills. They worked from caf\u00e9s, beaches, and co-living spaces\u2014delivering the same (or better) output than their office-bound peers.
Then, suddenly, the backlash came. CEOs demanded a Return to Office. Managers moralized about \u201ccollaboration.\u201d HR departments spun up surveillance tools to track badge swipes.
What happened?
The usual explanations\u2014productivity! culture! innovation!\u2014are weak. Study after study showed remote work didn\u2019t hurt performance. So why the crackdown?
The Girardian Unmasking
French philosopher Ren\u00e9 Girard\u2019s theories on mimetic desire and scapegoating explain this better than any MBA analysis. Here\u2019s how:
1. Mimetic Desire: The Office as a Shared Delusion
Girard argued that humans don\u2019t desire things directly\u2014we desire through others. For decades, white-collar workers mimicked a single model of success:
Live in an expensive city.
Grind 60-hour weeks.
Chase promotions for status.
This system worked because everyone bought into the same fiction. Then came remote work\u2014and a new model emerged:
Geoarbitrage.
Tax optimization.
Lifestyle freedom.
The moment workers saw colleagues thriving outside the system, the old model started crumbling. The desire shifted. But instead of adapting, corporations panicked.
2. The Scapegoat Mechanism: Sacrificing the Digital Nomad
When mimetic rivalry escalates, Girard says societies restore order by uniting against a scapegoat. Remote workers became that scapegoat.
Executives couldn\u2019t admit the truth (\u201cWe\u2019re jealous you escaped\u201d), so they invented moral crises:
\u201cRemote workers aren\u2019t collaborating!\u201d (Yet Slack/GitHub metrics disproved this.)
\u201cWe need office culture!\u201d (But only after commercial real estate values tanked.)
\u201cIt\u2019s not fair if some work remotely!\u201d (A classic Girardian complaint\u2014masking envy as justice.)
By forcing RTO, they ritually sacrificed the remote worker to restore the old order.
3. The Sacred Myth: \u201cOffices = Serious Business\u201d
Every institution has sacred myths it defends at all costs. For corporations, the myth is: \u201cReal work happens in offices.\u201d
Remote work exposed this as a lie. Worse, it revealed that:
Middle managers exist to supervise, not produce.
Commercial real estate is a paper tiger.
The tax-optimized nomad is winning capitalism better than the CEO.
The backlash wasn\u2019t about productivity\u2014it was about protecting the myth.
Today, hybrid work is a compromise no one loves\u2014the worst of both worlds. But the genie won\u2019t go back in the bottle. The more companies enforce RTO, the more they reveal the system\u2019s fragility.
If you have a scapegoat story, share it.","title":"How Remote Work Died: A Girardian Tragedy in Corporate America","updated_at":"2026-03-22T20:32:19Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cosmicbboy"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hi HN!
Niels here. I'm the creator of UnionML (https://github.com/unionai-oss/unionml), a Python MLOps framework that removes the boilerplate and friction associated with building and deploying machine learning systems to production.
I've been training and deploying models for almost a decade now, and one pain-point I've consistently had is managing the complexity of building and maintaining an ML stack that works for the entire model development lifecycle - from prototyping to production.
UnionML is built on top of Flyte (https://www.flyte.org) and exposes a functional interface for defining the building blocks of your ML application via decorators -- think Flask or FastAPI method endpoints -- and UnionML takes care of bundling them into microservices for different use cases such as:
- model training
- batch prediction
- online prediction
- (more coming soon!)
This project aims to unify the rich ecosystem of data, ML, and MLOps tools that have emerged over the last decade or so (e.g. MLFlow, Sagemaker, Spark, etc.) to provide a nice UX for model developers, in both individual and team settings."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: UnionML \u2013 a Python framework for building ML microservices"}},"_tags":["story","author_cosmicbboy","story_31825328","show_hn"],"author":"cosmicbboy","children":[31825364,31825395,31825432],"created_at":"2022-06-21T15:50:29Z","created_at_i":1655826629,"num_comments":3,"objectID":"31825328","points":12,"story_id":31825328,"story_text":"Hi HN!
Niels here. I'm the creator of UnionML (https://github.com/unionai-oss/unionml), a Python MLOps framework that removes the boilerplate and friction associated with building and deploying machine learning systems to production.
I've been training and deploying models for almost a decade now, and one pain-point I've consistently had is managing the complexity of building and maintaining an ML stack that works for the entire model development lifecycle - from prototyping to production.
UnionML is built on top of Flyte (https://www.flyte.org) and exposes a functional interface for defining the building blocks of your ML application via decorators -- think Flask or FastAPI method endpoints -- and UnionML takes care of bundling them into microservices for different use cases such as:
- model training
- batch prediction
- online prediction
- (more coming soon!)
This project aims to unify the rich ecosystem of data, ML, and MLOps tools that have emerged over the last decade or so (e.g. MLFlow, Sagemaker, Spark, etc.) to provide a nice UX for model developers, in both individual and team settings.","title":"Show HN: UnionML \u2013 a Python framework for building ML microservices","updated_at":"2024-09-20T11:24:03Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"xiaofei_"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hi HN - We're Ping-Lin and Xiaofei from Instill AI (https://www.instill.tech). We're building VDP (https://github.com/instill-ai/vdp), an open-source ETL tool for unstructured visual data.
When people say they are data-driven, most of the time it means they are driven by structured data. I will cut the part where we cite reports claiming that 80% of data are unstructured. The reality is unstructured data are more difficult to analyse and not a lot of companies know or have the resources to deal with them.
Before starting Instill AI, we were in a smart video startup dealing with large volumes of visual data every day. Back then, the concept of MLOps was pretty new (2014), every ML company was exploring and building its own stack. We built a battle-proven Vision AI system in-house and had the system running in production for years.
What we have learnt from the journey are: 1) buy vs. build: unaffordable high inference cost was the main barrier keeping us from adopting an off-the-shelf solution like Google Vertex AI or Amazon SageMaker, so we went for the "build" route. The truth was the resources we spent on building and maintaining the system were unexpectedly huge, time and money-wise; 2) the Vision AI system we built can actually be modularised and generalised to apply to other industry sectors.
We reckon what we had experienced can be a common phenomenon in the industry, and we can help solve the problem. That's why we decide to build VDP, an open-source, general and modularised ETL infrastructure for unstructured visual data for a broader community.
Many brilliant MLOps platforms/tools providing computer vision solutions have emerged in the last few years. Most of the tools are built from a model-centric perspective and fall into the following categories:
- general ML platforms for model training, experiment tracking, model deployment, etc.
- platforms that serve a specific vertical, such as E-commerce, and manufacturing.
- platforms that focus on a single component of MLOps, such as data labelling, dataset preparation, and model serving.
VDP is built from a data-driven perspective. Although the computer vision model is the most critical component in a visual data ETL pipeline, the ultimate goal of VDP is to streamline the end-to-end visual data flow, with the transform component being able to flexibly import computer vision models from different sources.
Today, the early version of VDP supports 2 sources and all Airbyte destination connectors, and it can import computer vision models from various sources including Local, GitHub, DVC, ArtiVC and Hugging Face. Setting up a VDP pipeline is fairly easy via its low-code API and no-code Console. Please take a look at the tutorial: https://www.instill.tech/docs/tutorials/build-an-async-det-p....
VDP can run locally with Docker Compose. We're working on integrating with Kubernetes and a fully managed version in Instill Cloud.
We aim to build VDP as the single point of visual data integration, so users can sync visual data from anywhere into centralised warehouses or applications and focus on gaining insights across all data sources, just like how the modern data stack handles structured data.
Operation-wise, VDP resources will be managed in a declarative way to fuse them better with the modern cloud-native context. The API-first and microservice design has opened all sorts of possibilities for VDP.
Thanks for reading HN! We are first-time open-source project maintainers. There are definitely lots to learn! Let us know what you think in the comments.
VDP links:
[1] GitHub: https://github.com/instill-ai/vdp
[2] Documentation: https://www.instill.tech/docs
[3] Demo: https://demo.instill.tech"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: VDP \u2013 open-source unstructured visual data ETL"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/instill-ai/vdp"}},"_tags":["story","author_xiaofei_","story_32546589","show_hn"],"author":"xiaofei_","children":[32549277],"created_at":"2022-08-22T02:17:55Z","created_at_i":1661134675,"num_comments":1,"objectID":"32546589","points":11,"story_id":32546589,"story_text":"Hi HN - We're Ping-Lin and Xiaofei from Instill AI (https://www.instill.tech). We're building VDP (https://github.com/instill-ai/vdp), an open-source ETL tool for unstructured visual data.
When people say they are data-driven, most of the time it means they are driven by structured data. I will cut the part where we cite reports claiming that 80% of data are unstructured. The reality is unstructured data are more difficult to analyse and not a lot of companies know or have the resources to deal with them.
Before starting Instill AI, we were in a smart video startup dealing with large volumes of visual data every day. Back then, the concept of MLOps was pretty new (2014), every ML company was exploring and building its own stack. We built a battle-proven Vision AI system in-house and had the system running in production for years.
What we have learnt from the journey are: 1) buy vs. build: unaffordable high inference cost was the main barrier keeping us from adopting an off-the-shelf solution like Google Vertex AI or Amazon SageMaker, so we went for the "build" route. The truth was the resources we spent on building and maintaining the system were unexpectedly huge, time and money-wise; 2) the Vision AI system we built can actually be modularised and generalised to apply to other industry sectors.
We reckon what we had experienced can be a common phenomenon in the industry, and we can help solve the problem. That's why we decide to build VDP, an open-source, general and modularised ETL infrastructure for unstructured visual data for a broader community.
Many brilliant MLOps platforms/tools providing computer vision solutions have emerged in the last few years. Most of the tools are built from a model-centric perspective and fall into the following categories:
- general ML platforms for model training, experiment tracking, model deployment, etc.
- platforms that serve a specific vertical, such as E-commerce, and manufacturing.
- platforms that focus on a single component of MLOps, such as data labelling, dataset preparation, and model serving.
VDP is built from a data-driven perspective. Although the computer vision model is the most critical component in a visual data ETL pipeline, the ultimate goal of VDP is to streamline the end-to-end visual data flow, with the transform component being able to flexibly import computer vision models from different sources.
Today, the early version of VDP supports 2 sources and all Airbyte destination connectors, and it can import computer vision models from various sources including Local, GitHub, DVC, ArtiVC and Hugging Face. Setting up a VDP pipeline is fairly easy via its low-code API and no-code Console. Please take a look at the tutorial: https://www.instill.tech/docs/tutorials/build-an-async-det-p....
VDP can run locally with Docker Compose. We're working on integrating with Kubernetes and a fully managed version in Instill Cloud.
We aim to build VDP as the single point of visual data integration, so users can sync visual data from anywhere into centralised warehouses or applications and focus on gaining insights across all data sources, just like how the modern data stack handles structured data.
Operation-wise, VDP resources will be managed in a declarative way to fuse them better with the modern cloud-native context. The API-first and microservice design has opened all sorts of possibilities for VDP.
Thanks for reading HN! We are first-time open-source project maintainers. There are definitely lots to learn! Let us know what you think in the comments.
VDP links:
[1] GitHub: https://github.com/instill-ai/vdp
[2] Documentation: https://www.instill.tech/docs
[3] Demo: https://demo.instill.tech","title":"Show HN: VDP \u2013 open-source unstructured visual data ETL","updated_at":"2024-09-20T11:51:20Z","url":"https://github.com/instill-ai/vdp"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"navaneethpk"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"I\u2019m excited to share ToolJet 3.0, a major update to our open-core low-code platform for building internal tools. This release is all about speed, scalability, and better support for complex use cases.
Backstory
When I first launched ToolJet (https://github.com/ToolJet/ToolJet) here on HN (https://news.ycombinator.com/item?id=27421408](https://news.... 2021, it started as a tool for building frontends connected to data sources via a secure backend proxy, with integrations like Google Sheets. The initial post hit #1 on HN, bringing in a wave of early adopters and helping us secure seed funding.
Since then, ToolJet has grown significantly. We added a visual workflow builder, a no-code database on PostgreSQL, and expanded integrations. But as the platform evolved, the codebase became heavier, and performance issues emerged. ToolJet 3.0 is a complete refactor focused on fixing these problems and delivering a faster, more reliable experience.
What\u2019s New in ToolJet 3.0
Here\u2019s what we\u2019ve built for this release:
\u202210x Faster Load Times: We refactored the core architecture, resulting in much faster performance, especially for complex apps.\n \u2022Enhanced Workflow Automation: The visual builder now handles complex workflows with better execution speed and reliability.\n \u2022Upgraded No-Code Database: Built on PostgreSQL, our database now scales better and supports more data types.\n \u2022Expanded Integrations: Added Salesforce, Supabase, Databricks, AWS Textract, OpenAI, Pinecone, and others based on user feedback.\n \u2022Security and Access Control: New features like LDAP, SAML, Git sync, and OIDC group sync provide enterprise-level access control.\n \u2022More Self-Hosting Options: We\u2019ve expanded support for different platforms, making it easier to deploy ToolJet in your own environment.\n \u2022UI and UX Improvements: Redesigned form components, added a Kanban board, and made several enhancements based on user feedback.\n \u2022ToolJet Copilot: An AI-powered assistant for query writing and debugging, aimed at making development faster and more intuitive.\n \u2022Community Growth: Our GitHub repo has surpassed 32,000 stars with contributions from over 600 developers.
Why It Matters
With 3.0, we\u2019re doubling down on our vision of a fast, flexible, and extensible platform that balances the power of an open-core model with the needs of modern development teams. This release addresses the stability and scalability issues that came with rapid growth, making ToolJet ready for complex internal tools."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: ToolJet 3.0 \u2013 open-source internal tool and workflow builder"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/ToolJet/ToolJet"}},"_tags":["story","author_navaneethpk","story_42130699","show_hn"],"author":"navaneethpk","created_at":"2024-11-13T22:13:19Z","created_at_i":1731535999,"num_comments":0,"objectID":"42130699","points":11,"story_id":42130699,"story_text":"I\u2019m excited to share ToolJet 3.0, a major update to our open-core low-code platform for building internal tools. This release is all about speed, scalability, and better support for complex use cases.
Backstory
When I first launched ToolJet (https://github.com/ToolJet/ToolJet) here on HN (https://news.ycombinator.com/item?id=27421408](https://news.... 2021, it started as a tool for building frontends connected to data sources via a secure backend proxy, with integrations like Google Sheets. The initial post hit #1 on HN, bringing in a wave of early adopters and helping us secure seed funding.
Since then, ToolJet has grown significantly. We added a visual workflow builder, a no-code database on PostgreSQL, and expanded integrations. But as the platform evolved, the codebase became heavier, and performance issues emerged. ToolJet 3.0 is a complete refactor focused on fixing these problems and delivering a faster, more reliable experience.
What\u2019s New in ToolJet 3.0
Here\u2019s what we\u2019ve built for this release:
\u202210x Faster Load Times: We refactored the core architecture, resulting in much faster performance, especially for complex apps.\n \u2022Enhanced Workflow Automation: The visual builder now handles complex workflows with better execution speed and reliability.\n \u2022Upgraded No-Code Database: Built on PostgreSQL, our database now scales better and supports more data types.\n \u2022Expanded Integrations: Added Salesforce, Supabase, Databricks, AWS Textract, OpenAI, Pinecone, and others based on user feedback.\n \u2022Security and Access Control: New features like LDAP, SAML, Git sync, and OIDC group sync provide enterprise-level access control.\n \u2022More Self-Hosting Options: We\u2019ve expanded support for different platforms, making it easier to deploy ToolJet in your own environment.\n \u2022UI and UX Improvements: Redesigned form components, added a Kanban board, and made several enhancements based on user feedback.\n \u2022ToolJet Copilot: An AI-powered assistant for query writing and debugging, aimed at making development faster and more intuitive.\n \u2022Community Growth: Our GitHub repo has surpassed 32,000 stars with contributions from over 600 developers.
Why It Matters
With 3.0, we\u2019re doubling down on our vision of a fast, flexible, and extensible platform that balances the power of an open-core model with the needs of modern development teams. This release addresses the stability and scalability issues that came with rapid growth, making ToolJet ready for complex internal tools.","title":"Show HN: ToolJet 3.0 \u2013 open-source internal tool and workflow builder","updated_at":"2024-11-14T20:56:05Z","url":"https://github.com/ToolJet/ToolJet"}],"hitsPerPage":50,"nbHits":70,"nbPages":2,"page":0,"params":"query=Emerald+AI&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":10,"processingTimingsMS":{"_request":{"queue":11,"roundTrip":21},"afterFetch":{"format":{"highlighting":5,"total":6},"merge":{"mergeLoop":{"prepareNextHit":1,"total":1},"total":1},"total":1},"fetch":{"query":5,"scanning":2,"total":8},"total":10},"query":"Emerald AI","serverTimeMS":28}