I built AIDictation.com, a voice to text app written in Swift. It sends audio to my own backend, runs it through a Whisper-based pipeline, and returns a transcription you can then send straight into an AI chat like ChatGPT or Claude.
I\u2019ve been building full\u2011stack apps for ~20 years, but this is my first Swift application. I leaned heavily on AI coding tools to get from zero Swift to a working app and backend in a couple of weeks.
What it does
Records audio and sends it to my server.\nThe backend runs a pipeline using Whisper V3 Turbo + OpenAI GPT OSS 120B.
I intentionally went with a cloud pipeline instead of on\u2011device models so I can:\n- Parallelize work on the backend and tune the pipeline.\n- Mix and match providers and models.\n- Improve latency without shipping new app versions.
After transcription, there\u2019s a \u201cshare to AI chat\u201d flow so you can send it with one tap to ChatGPT, Claude, etc.
Context rules\nOne feature I missed in Whisper Flow was configurable context rules (similar to the Super Whisper Modes). AIDictation lets you define how transcription should behave depending on what you\u2019re doing.
For example:\n- Meetings: keep speaker names and timestamps.\n- Coding: preserve technical terms and code formatting.\n- Journaling: be more forgiving, add punctuation, make the text more readable.\n- You can configure different presets and switch between them.
Why cloud instead of on\u2011device
A lot of apps focus on running models locally. I chose the opposite trade\u2011off:\n- Provider flexibility: right now I\u2019m using the Groq API because, in my tests, it had the best end\u2011to\u2011end latency (700-800ms), but the backend is built to swap providers and models.\n- This does mean audio leaves the device, so I tried to be very explicit about data handling.
No registration needed. You get about 2,000 words per month for free without creating an account or giving an email.
Tech stack\nClient: Swift (first real Swift/iOS app I\u2019ve shipped).\nBackend: NodeJS on Vercel.\nModels: Whisper V3 Turbo + OpenAI GPT OSS 120B.\nProvider: Groq API at the moment, mainly for latency reasons.
I\u2019ve been using AIDictation daily for the past couple of weeks, and I\u2019m happy with it so far, but I\u2019d really like candid feedback from HN\u2014both on the product and on the implementation."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: AIDictation \u2013 zero data retention dictation app"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://aidictation.com"}},"_tags":["story","author_vood","story_46094955","show_hn"],"author":"vood","created_at":"2025-11-30T08:40:30Z","created_at_i":1764492030,"num_comments":0,"objectID":"46094955","points":2,"story_id":46094955,"story_text":"Hi HN,
I built AIDictation.com, a voice to text app written in Swift. It sends audio to my own backend, runs it through a Whisper-based pipeline, and returns a transcription you can then send straight into an AI chat like ChatGPT or Claude.
I\u2019ve been building full\u2011stack apps for ~20 years, but this is my first Swift application. I leaned heavily on AI coding tools to get from zero Swift to a working app and backend in a couple of weeks.
What it does
Records audio and sends it to my server.\nThe backend runs a pipeline using Whisper V3 Turbo + OpenAI GPT OSS 120B.
I intentionally went with a cloud pipeline instead of on\u2011device models so I can:\n- Parallelize work on the backend and tune the pipeline.\n- Mix and match providers and models.\n- Improve latency without shipping new app versions.
After transcription, there\u2019s a \u201cshare to AI chat\u201d flow so you can send it with one tap to ChatGPT, Claude, etc.
Context rules\nOne feature I missed in Whisper Flow was configurable context rules (similar to the Super Whisper Modes). AIDictation lets you define how transcription should behave depending on what you\u2019re doing.
For example:\n- Meetings: keep speaker names and timestamps.\n- Coding: preserve technical terms and code formatting.\n- Journaling: be more forgiving, add punctuation, make the text more readable.\n- You can configure different presets and switch between them.
Why cloud instead of on\u2011device
A lot of apps focus on running models locally. I chose the opposite trade\u2011off:\n- Provider flexibility: right now I\u2019m using the Groq API because, in my tests, it had the best end\u2011to\u2011end latency (700-800ms), but the backend is built to swap providers and models.\n- This does mean audio leaves the device, so I tried to be very explicit about data handling.
No registration needed. You get about 2,000 words per month for free without creating an account or giving an email.
Tech stack\nClient: Swift (first real Swift/iOS app I\u2019ve shipped).\nBackend: NodeJS on Vercel.\nModels: Whisper V3 Turbo + OpenAI GPT OSS 120B.\nProvider: Groq API at the moment, mainly for latency reasons.
I\u2019ve been using AIDictation daily for the past couple of weeks, and I\u2019m happy with it so far, but I\u2019d really like candid feedback from HN\u2014both on the product and on the implementation.","title":"Show HN: AIDictation \u2013 zero data retention dictation app","updated_at":"2026-03-05T23:10:38Z","url":"https://aidictation.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"leopiney"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["whisperflow"],"value":"Just vibe-coded this app today, hope it works for you if don't want to pay SuperWhisper/WhisperFlow/Monologue apps and happen to have an ElevenLabs subscription."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: macOS ElevenLabs Scribe v2 app"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/leopiney/elevenscribe"}},"_tags":["story","author_leopiney","story_47405064","show_hn"],"author":"leopiney","created_at":"2026-03-16T21:21:50Z","created_at_i":1773696110,"num_comments":0,"objectID":"47405064","points":1,"story_id":47405064,"story_text":"Just vibe-coded this app today, hope it works for you if don't want to pay SuperWhisper/WhisperFlow/Monologue apps and happen to have an ElevenLabs subscription.","title":"Show HN: macOS ElevenLabs Scribe v2 app","updated_at":"2026-03-16T21:23:24Z","url":"https://github.com/leopiney/elevenscribe"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"nirajjha"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["whisperflow"],"value":"I recently explored Vibe Coding, a new approach to building apps using voice and AI, and wanted to share my experience.
It\u2019s an exciting way to develop without writing every line of code manually. Here\u2019s a quick breakdown:
Vibe Coding Starter Pack: Build Apps by Talking to AI
What is Vibe Coding?\nVibe Coding allows you to create apps by speaking to AI tools like Cursor and Whisper Flow, which convert your ideas into working code. This approach makes development accessible, intuitive, and efficient for both beginners and experienced developers.
How It Works:\nSay your idea \u2192 AI writes the code.\nCursor handles coding; Whisper Flow converts speech to text.\nNo coding experience? No problem\u2014just describe what you want.
Benefits of Vibe Coding\nEffortless development: Voice replaces manual coding.\nFlexible and fast: Iterate with real-time feedback.\nNo syntax struggles: AI does the heavy lifting.\nSecurity matters: Store API keys securely and document everything.
Tools for Vibe Coding\nAI Coding Assistants\nCursor \u2013 AI-powered coding assistant.\nWhisper Flow \u2013 Speech-to-text tool.\nSuper Whisper \u2013 High-performance voice-to-code transcription.
Open-Source Alternatives\nAide \u2013 Lightweight AI coding assistant.\nDSNote \u2013 AI-powered note-taking tool.\nSay \u2013 Open-source speech-to-text for coding.\nSpeech Transcription Tools
For more tools, check out Awesome Whisper.
Curious about Vibe Coding? Explore these tools and start turning your voice into apps today!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Vibe Coding Starter Pack: Build Apps by Talking to AI"}},"_tags":["story","author_nirajjha","story_43025811","ask_hn"],"author":"nirajjha","children":[43025819],"created_at":"2025-02-12T14:38:18Z","created_at_i":1739371098,"num_comments":0,"objectID":"43025811","points":1,"story_id":43025811,"story_text":"I recently explored Vibe Coding, a new approach to building apps using voice and AI, and wanted to share my experience.
It\u2019s an exciting way to develop without writing every line of code manually. Here\u2019s a quick breakdown:
Vibe Coding Starter Pack: Build Apps by Talking to AI
What is Vibe Coding?\nVibe Coding allows you to create apps by speaking to AI tools like Cursor and Whisper Flow, which convert your ideas into working code. This approach makes development accessible, intuitive, and efficient for both beginners and experienced developers.
How It Works:\nSay your idea \u2192 AI writes the code.\nCursor handles coding; Whisper Flow converts speech to text.\nNo coding experience? No problem\u2014just describe what you want.
Benefits of Vibe Coding\nEffortless development: Voice replaces manual coding.\nFlexible and fast: Iterate with real-time feedback.\nNo syntax struggles: AI does the heavy lifting.\nSecurity matters: Store API keys securely and document everything.
Tools for Vibe Coding\nAI Coding Assistants\nCursor \u2013 AI-powered coding assistant.\nWhisper Flow \u2013 Speech-to-text tool.\nSuper Whisper \u2013 High-performance voice-to-code transcription.
Open-Source Alternatives\nAide \u2013 Lightweight AI coding assistant.\nDSNote \u2013 AI-powered note-taking tool.\nSay \u2013 Open-source speech-to-text for coding.\nSpeech Transcription Tools
For more tools, check out Awesome Whisper.
Curious about Vibe Coding? Explore these tools and start turning your voice into apps today!","title":"Vibe Coding Starter Pack: Build Apps by Talking to AI","updated_at":"2025-02-12T14:42:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"js98"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["whisperflow"],"value":"Looking at tech blogs, HN submissions, et cetera, a lot of content is (partly, increasingly?) generated by AI. The value of the reader's time and slop has been discussed at length. I ask:
Among you, the HN-community who writes, how much do you use AI in the process?
There is somewhat a spectrum from none, to grammarly-style cleanup, to whisprflow to 'look at my project, write me a HN-ready post'.
Personally, I've done a 180, where I started at light touchups to rough draft-to-polished back to almost none, to keep the writing my own (I do some writing at jakobs.dev).
As a reader, do you believe you notice, and care?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: How much AI is in your writing?"}},"_tags":["story","author_js98","story_48235223","ask_hn"],"author":"js98","children":[48235280,48235297,48235302,48235497,48235521,48235535,48236259,48236462,48236488,48236669,48237034,48237080,48237344,48237392,48237397,48238456,48239028,48239956,48239970,48240485,48241186,48241204,48241333,48241399,48241777,48245563,48245697,48245714,48255034,48255726,48261408,48261459,48261471,48263145,48263298,48263381,48268090,48272977,48278129],"created_at":"2026-05-22T12:55:35Z","created_at_i":1779454535,"num_comments":37,"objectID":"48235223","points":10,"story_id":48235223,"story_text":"Looking at tech blogs, HN submissions, et cetera, a lot of content is (partly, increasingly?) generated by AI. The value of the reader's time and slop has been discussed at length. I ask:
Among you, the HN-community who writes, how much do you use AI in the process?
There is somewhat a spectrum from none, to grammarly-style cleanup, to whisprflow to 'look at my project, write me a HN-ready post'.
Personally, I've done a 180, where I started at light touchups to rough draft-to-polished back to almost none, to keep the writing my own (I do some writing at jakobs.dev).
As a reader, do you believe you notice, and care?","title":"Ask HN: How much AI is in your writing?","updated_at":"2026-06-05T11:39:04Z"}],"hitsPerPage":50,"nbHits":9,"nbPages":1,"page":0,"params":"query=Whisperflow&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":7,"processingTimingsMS":{"_request":{"roundTrip":20},"afterFetch":{"merge":{"total":1},"total":1},"fetch":{"query":4,"total":5},"total":7},"query":"Whisperflow","serverTimeMS":8}