Codebase Indexing

Mita can index your codebase for retrieval-augmented generation (RAG). When enabled, the agent automatically retrieves relevant code snippets to provide better, context-aware responses.

How It Works

1. Parse — Tree-sitter extracts semantic chunks (functions, classes) from your code
2. Embed — Ollama generates vector embeddings using nomic-embed-text
3. Store — Chunks are stored in a local LanceDB database (.mita/index/)
4. Retrieve — When you ask a question, relevant chunks are injected into the LLM context

Build the Index

mita index build

If the embedding model isn't installed, Mita will prompt you to pull it.

Use --force to rebuild from scratch:

mita index build --force

Search

mita index search "database connection pooling"
mita index search "authentication" --top-k 5

Status

mita index status

Shows chunk count, index size, and last build time.

Clear

mita index clear

Deletes the index. Rebuild with mita index build.

Configuration

[index]
enabled = true
chunk_size = 512
chunk_overlap = 64
top_k = 10
exclude_patterns = [
    "*.lock",
    ".mita/**",
    "node_modules/**",
    ".git/**",
    "*.min.js",
    "*.min.css",
    "dist/**",
    "build/**",
    "__pycache__/**",
]

Supported Languages

Tree-sitter parsing supports semantic chunking for: Python, JavaScript, TypeScript, Rust, Go, Java, C, C++, Ruby, PHP, and more. Files in unsupported languages fall back to line-based chunking.