← all repositories
akshayaggarwal99/jarvis-ai-assistant

One dev's $81 million spite project

A free, open-source voice dictation app for Mac built to prove you don't need VC funding or subscriptions for decent AI tooling.

571 stars TypeScript AgentsChat Assistants
jarvis-ai-assistant
Not currently ranked — collecting fresh signals.
star history

What it does

Jarvis is a Mac-native voice assistant that transcribes, cleans up, and posts your speech anywhere you can type. Hold the Fn key, ramble, release — it strips filler words, fixes grammar, and can rephrase or bullet-point the result. It also handles small commands like “set 5-min timer” or “open YouTube.” The whole thing runs locally if you want it to, or you can plug in cloud APIs for speed.

The interesting bit

The origin story is doing a lot of heavy lifting here: the author explicitly built this after Wispr Flow raised $81 million for what he saw as a simple dictation app. Three months of late nights, zero team, MIT license. The README’s comparison table pitting “1 guy at 2am” against “$700M startups” is either refreshingly honest or slightly unhinged — possibly both. The local-first architecture is genuinely notable: Whisper or NVIDIA Parakeet via Sherpa-ONNX for transcription, Ollama for LLM post-processing, with Deepgram + Gemini as optional cloud fallbacks.

Key highlights

  • Fully offline capable with local Whisper/Parakeet transcription and Ollama LLM support
  • Apple-signed and notarized; no gatekeeper warnings on install
  • Custom prompt engineering for dictation style, email formatting, assistant behavior
  • iOS TestFlight version exists (separate from Mac app)
  • Deepgram requests default to mip_opt_out=true — actual privacy-by-default, not just marketing
  • Double-tap Fn for hands-free mode; Escape cancels mid-recording

Caveats

  • Mac-only; Windows is on the roadmap but doesn’t exist yet
  • Local LLM post-processing adds ~1–3s latency on standard M1/M2 chips; the README itself recommends cloud for “lightning fast” use
  • Some roadmap items (streaming transcription, clipboard magic, multi-step actions) are still pending

Verdict

Worth a look if you’re a Mac user who dictates regularly and resents subscription pricing for basic utilities. Skip it if you need Windows, real-time streaming transcription, or aren’t willing to trade some latency for privacy. The project wears its anti-VC ethos loudly — that resonates or grates depending on your tolerance for manifestos in READMEs.

Frequently asked

What is akshayaggarwal99/jarvis-ai-assistant?
A free, open-source voice dictation app for Mac built to prove you don't need VC funding or subscriptions for decent AI tooling.
Is jarvis-ai-assistant open source?
Yes — akshayaggarwal99/jarvis-ai-assistant is open source, released under the MIT license.
What language is jarvis-ai-assistant written in?
akshayaggarwal99/jarvis-ai-assistant is primarily written in TypeScript.
How popular is jarvis-ai-assistant?
akshayaggarwal99/jarvis-ai-assistant has 571 stars on GitHub.
Where can I find jarvis-ai-assistant?
akshayaggarwal99/jarvis-ai-assistant is on GitHub at https://github.com/akshayaggarwal99/jarvis-ai-assistant.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.