Repository-aware context · Controlled execution · Your models
Warning
Website and docs are in progress.
Please wait for the public release — or, if you’re curious now, use module-specific READMEs inside the monorepo (e.g. packages/v8/, apps/vscode/, packages/sdk/).
mitii.dev and docs.mitii.dev are not the source of truth yet.
Mitii is a local-first AI coding agent for VS Code. It indexes your repository, reasons with bounded context, and applies changes under approvals and checkpoints — with local models (Ollama, LM Studio) or any OpenAI-compatible / Anthropic / Gemini endpoint.
Not “chat that dumps a patch.” A short, inspectable loop:
[mitii] 1. MEASURE ➔ Indexed 128 files | Scope: 14k tokens
[mitii] 2. IDENTIFY ➔ Symbol resolution complete | Root cause pinned to auth.ts:42
[mitii] 3. TASK ➔ Drafted Plan/Act contract | 3 files targeted for edit
[mitii] 4. IMPLEMENT ➔ Patch applied | Git checkpoint saved (stash@0)
[mitii] 5. IMPROVE ➔ TypeScript build clean | 12/12 unit tests passing
| Product | Mitii — monorepo (VS Code extension, CLI, V8 runtime, SDK) |
| Install | VS Code Marketplace — Mitii AI Agent |
| Architecture | packages/v8/ARCHITECTURE.md |
| Extension | apps/vscode/README.md |
| Issues | Mitii-dev/Mitii/issues |
We build agent infrastructure you can inspect: hybrid retrieval, decision policy, tool runtime, verification — host-neutral runtime (@mitii/v8) behind thin hosts (VS Code, CLI).
VS Code / CLI → @mitii/sdk → @mitii/v8 → models + tools + index
Privacy-first by default. Run fully local when you want; cloud when you need more capacity.