Offline-first AI for VS Code. Code completions, chat, and code actions. No API keys. No cloud required.
Features
Chat
Ask questions, generate code, debug errors. Streaming responses in a side panel.
Inline Completions
Type and get suggestions as you code. Accept with Tab.
Code Actions (Right-click)
- Explain Selection — understand unfamiliar code
- Fix Selection — get a corrected version with explanation
Multilingual
Responds in your language. Works in English, Portuguese, Spanish, and others.
Not a cloud AI
- No OpenAI API
- No remote code execution
- No data sent to third-party servers
- No account or login required
Runs locally by default. Network features are optional and never required.
Install
- Install "Choir Code" from the VS Code Marketplace
- Wait ~60 seconds (downloads runtime + model on first use)
- Start coding
No Ollama. No Docker. No config files.
Performance
| Metric | Value |
|---|---|
| First load | ~30–60s (model download + warmup) |
| First response | ~2–5s |
| Subsequent responses | ~1–2s |
| RAM usage | ~600MB during inference |
| Disk | ~600MB (runtime + model) |
Requirements
- Any x86_64 CPU (Intel/AMD, 2011 or newer)
- ~600MB free disk
- ~600MB free RAM
- Works fully offline
How it works
Choir runs a small AI model (0.5B parameters) directly on your machine. Your code never leaves your computer.
Optional network verification when other nodes are online improving confidence in responses. This is automatic and never required.
Settings
| Setting | Default | Description |
|---|---|---|
choir.endpoint |
http://localhost:5001 |
Local node address |
Troubleshooting
Chat not responding after install?
- Wait 60s (model is downloading on first use)
- Check Output panel → "Choir" for errors
Slow first response?
- Model loads into RAM on first request (~2-5s)
- Subsequent responses are fast (~1-2s)
High memory usage?
- The model uses ~600MB RAM while active
- Closing VS Code frees the memory
Links
"Every cycle of computation produces real intelligence. Nothing is wasted."