Mita Code¶
Local-first, terminal-native agentic coding assistant powered by Ollama.
No API keys. No cloud. No telemetry. Your code never leaves your machine.
What is Mita Code?¶
Mita Code (mita) is a CLI coding assistant that runs large language models entirely on your hardware via Ollama. It gives you an agentic tool loop — reading files, writing code, running shell commands, searching your codebase — all from the terminal, all locally.
Key Features¶
- :material-server-network: **100% Local** — All inference runs on your hardware via Ollama.
- :material-tools: **Agentic Tool Loop** — Read/write files, run shell commands, git operations with confirmation for destructive actions.
- :material-memory: **Layered Memory** — `MITA.md` files at global, project, and directory scope are automatically injected into context.
- :material-magnify: **Codebase Indexing** — Local vector search (LanceDB + Tree-sitter) for RAG over your codebase.
- :material-lightning-bolt: **Skills** — Reusable prompt templates stored as Markdown (e.g., `/commit`, `/review`).
- :material-chip: **Hardware-Aware** — Detects your RAM, VRAM, and GPU to recommend models that run well on your machine.
- :material-puzzle: **MCP Plugins** — Compatible with the [Model Context Protocol](https://modelcontextprotocol.io/) ecosystem.
- :material-cog: **Layered Config** — TOML configuration cascades from global to project level.
Quick Start¶
# Install
pipx install mita-code
# Start Ollama
ollama serve
# Pull a coding model
mita models pull qwen2.5-coder:7b
# Start an interactive session
mita chat
See the Installation and Quick Start guides for details.
Tech Stack¶
| Component | Library |
|---|---|
| CLI | Typer |
| Terminal UI | Rich |
| LLM Runtime | Ollama |
| LLM Client | LiteLLM |
| Vector Store | LanceDB |
| Code Parsing | Tree-sitter |
| Config | TOML (stdlib tomllib) |
| Plugins | MCP |