Jarvis AI Assistant

Hold one key. Speak. Text appears — perfectly.

100% open-source • 100% local-capable • 100% free forever

https://github.com/user-attachments/assets/763ea6aa-87d5-4e0a-9f05-a2bdaa82f40f

▶ Watch full 2-min demo on YouTube

---

The (short) story

Wispr Flow raised $81 million to build a voice dictation app.
I got annoyed, spent 3 months of late nights building my own version, and open-sourced it.

That's it.
No funding. No team. Just one stubborn developer who hates subscriptions.

This is what open source is supposed to be.

---

Why people actually use it

• Hold Fn → speak → release → clean, punctuated text appears anywhere
• Removes "um", "like", all fillers automatically
• Fixes grammar, can rephrase, bullet-point, or even generate text
• Tiny actions already work ("open YouTube", "set 5-min timer")
• Full Prompt Engineering: Every behavior is customizable. Tweak how Jarvis formats mail, cleans dictation, or behaves as an assistant.
• Fully offline with local Whisper or NVIDIA Parakeet (via Sherpa-ONNX)
• 100% Private, Blazingly Fast support for local LLMs via Ollama
• Or use cloud speed with Deepgram + Gemini (1M tokens/day free) — Deepgram requests default to mip_opt_out=true

Zero tracking. Zero telemetry. Zero bullshit.

---

Download

✅ Signed & notarized by Apple → no scary warnings.

---

Jarvis vs $700M startups

---

Quick setup (30 seconds)

1. Download & open the app
2. Go Local (Recommended for privacy):
   • Settings → Transcription → Enable Local Model
   • Choose Whisper (tiny/base/small) or Parakeet (NVIDIA's 600M param model)
   • Settings → AI Models → Enable Ollama (requires Ollama installed)
3. Or Go Cloud (Recommended for speed/accuracy):
   • Paste free Deepgram + Gemini keys
4. Hold Fn and talk

That's literally it.

---

🦙 Ollama (Local LLM) Support

Jarvis now supports running any LLM locally via Ollama. It's fast, private, and free.

1. Install Ollama: Download here
2. Pull a model:

   # LFM2 (optimized for speed)
   ollama pull sam860/LFM2:1.2b
   
   # Llama 3 (standard)
   ollama pull llama3
   

3. Enable in Jarvis:
   • Settings → AI Models → Enable Use Ollama
   • Select your model from the dropdown (Jarvis auto-detects them!)

⚡ Performance Note

Running models locally depends heavily on your hardware.

• Enabled AI Post-Processing: Adds ~1-3s latency on typical M1/M2 chips.
• Disabled AI Post-Processing: Instant transcription.
• High-End Hardware: M1/M2/M3 Max/Ultra chips are nearly instant.

For Lightning Fast Performance: If speed is critical, use Deepgram + Gemini (Cloud) or Local Whisper + Gemini Flash, which are significantly faster than local LLMs on standard hardware.

---

Keyboard shortcuts

---

Build from source

git clone https://github.com/akshayaggarwal99/jarvis-ai-assistant.git
cd jarvis-ai-assistant
npm install && npm run build && npm run dev

Requires Node.js 18+, macOS 10.13+, Xcode CLI tools.

---

Roadmap (help me choose!)

• Proper "local-only" onboarding (no fake keys needed)
• AWS Bedrock support
• Ollama support
• Custom voice commands & prompts
• iOS (TestFlight available!)
• NVIDIA Parakeet support (via Sherpa-ONNX) — blazing fast local transcription
• Windows version
• Clipboard magic & multi-step actions
• Streaming transcription (real-time words as you speak)

Open issues, vote, or drop crazy ideas → github.com/.../issues

---

Wanna contribute?

Jarvis is a community project. If you're a developer and want to help build the future of local-first AI, check out our Contributing Guidelines.

Wanna help?

Every star pushes this higher so more people discover they don't need another subscription.

     

Or just use it and tell one friend. That's enough.

---

⭐ Star History

---

Built with caffeine and spite by Akshay

Open source isn't about beating giants.
It's about making sure no one ever has to pay them.

_{MIT License — do whatever you want with it.}