ElytraSync — Local AI for VS Code & Cursor
100% offline. Zero API cost. Your code stays on your machine.
ElytraSync brings a local AI coding assistant powered by elytra-model-mini directly into your editor — no cloud, no subscription, no data leaving your computer.
Features
Talk to the model directly in the sidebar. It automatically picks up the context of the file you have open, so you can ask things like "explain this function", "what's wrong here?" or "how do I refactor this?" — and get instant, relevant answers.
⚡ Inline Autocomplete (FIM)
Get Copilot-style code completions as you type. The model uses Fill-in-the-Middle technique — it sees both what's before and after your cursor — to suggest more accurate completions. Press Tab to accept.
🔍 CodeLens Alerts
ElytraSync automatically detects complex functions (high nesting, long bodies) and shows a CodeLens action above them. One click opens a refactor suggestion.
✨ Refactor with Diff
Select any code block → right-click → ElytraSync: Refatorar função. A side-by-side diff appears. Review the suggestion and choose Apply or Discard.
🧪 Generate Unit Tests
Select a function → ElytraSync: Gerar testes. The model generates happy-path, edge cases, and error tests directly in the chat.
📖 Explain Function
Select any function → ElytraSync: Explicar função. Get a plain-language, step-by-step explanation in the chat.
Requirements
Ollama — the local model runtime. Install and run:
ollama serve
elytra-model-mini — downloaded automatically on first launch via the onboarding screen.
On first install, ElytraSync opens a setup screen that checks Ollama and downloads the model for you with a progress bar.
Quick Start
- Install the extension
- The Welcome screen opens automatically
- If Ollama is missing → click Download Ollama → install → come back
- Click Download model → wait for the progress bar to complete
- Click Open Chat → start coding with AI
Keyboard Shortcuts
| Action |
Mac |
Windows/Linux |
| Open Chat |
Cmd+Shift+I C |
Ctrl+Shift+I C |
| Refactor selection |
Cmd+Shift+I R |
Ctrl+Shift+I R |
| Generate tests |
Cmd+Shift+I T |
Ctrl+Shift+I T |
Settings
| Setting |
Default |
Description |
iris.ollamaUrl |
http://localhost:11434 |
Ollama server URL |
iris.model |
qwen2.5-coder:3b |
Underlying Ollama model |
iris.autocomplete.enabled |
true |
Enable/disable inline autocomplete |
iris.autocomplete.debounceMs |
300 |
Debounce delay in ms |
iris.autocomplete.maxPrefixChars |
1500 |
Max context chars sent to the model |
iris.codeLens.enabled |
true |
Show CodeLens on complex functions |
iris.codeLens.minLines |
15 |
Min lines to trigger CodeLens |
iris.telemetry.enabled |
true |
Record usage locally for model training |
Privacy note: When iris.telemetry.enabled is true, your interactions (code snippets and AI responses) are saved locally in a JSONL file on your machine. This data is used to improve future versions of elytra-model-mini. Nothing is sent to any server. You can disable this at any time.
Commands (Command Palette)
| Command |
Description |
ElytraSync: Abrir chat |
Open the AI chat sidebar |
ElytraSync: Refatorar função |
Refactor selected code |
ElytraSync: Gerar testes |
Generate unit tests |
ElytraSync: Explicar função |
Explain selected function |
ElytraSync: Verificar status |
Check Ollama + model status |
ElytraSync: Ver estatísticas do dataset |
View local training dataset stats |
ElytraSync: Resetar onboarding (dev) |
Re-show the welcome screen |
Privacy & Data
- All AI inference runs locally via Ollama — no data leaves your machine
- Dataset collection (when enabled) writes to
<vscode-globalStorage>/dataset.jsonl — local only
- No telemetry is sent to ElytraSync servers
- Disable collection: set
iris.telemetry.enabled: false in settings
About ElytraSync
ElytraSync is building elytra-model-mini — a code model trained on real developer interactions, designed to get smarter over time. The extension is free. The model is local. The data you generate (with your consent) helps train the next version.
Website
