FAQ¶

General¶

What languages does repobrain support?¶

Python, TypeScript, JavaScript, Go, Rust, Java. Additional languages fall back to regex-based import detection.

Does repobrain send my code to Anthropic?¶

Yes — file content is sent to the Anthropic API for embedding and documentation generation. If this is a concern:

Use --no-docs to skip doc generation entirely (no file content sent for generation, only for embeddings)
Self-host a compatible embedding model and set REPOMIND_EMBEDDING_MODEL
Review Anthropic's data privacy policy

How much does it cost to index a typical repo?¶

A 1,000-file Python repo with claude-sonnet-4-6 for docs typically costs $0.80–$1.50 for the initial full index. Incremental updates cost a fraction of that — only changed files are re-processed.

Use --no-docs to eliminate LLM costs entirely if you only want the graph + metrics.

Can I use a different model?¶

Yes:

REPOMIND_LLM_MODEL=claude-haiku-4-5-20251001 repobrain index

Haiku is ~10× cheaper at some quality reduction. Opus gives best results at higher cost.

Installation¶

Does repobrain require Docker?¶

No. It runs entirely locally — SQLite, LanceDB, and NetworkX are all embedded. No external services required beyond the Anthropic API.

Can I use repobrain offline?¶

Partially. The graph, metrics, and SQL data are local. MCP tools that don't require LLM calls (get_hotspots, get_dependencies, get_cochange_patterns) work offline. Anything requiring generation or embedding needs Anthropic API access.

Does it work on Windows?¶

It runs on Windows via WSL2. Native Windows support is experimental.

MCP & Claude Code¶

How do I add repobrain to Claude Code?¶

repobrain serve --mcp-only

Then in Claude Code settings, add:

{
  "mcpServers": {
    "repobrain": {
      "command": "repobrain",
      "args": ["serve", "--mcp-only"]
    }
  }
}

Does repobrain work with other MCP clients?¶

Yes — it uses fastmcp which is compatible with any MCP client (Claude Code, Claude Desktop, custom clients).

How many repos can I index?¶

Each repo gets its own isolated data directory under REPOMIND_DATA_DIR. You can index as many repos as your disk allows. Switch between them:

repobrain serve --repo /path/to/repo-a
repobrain serve --repo /path/to/repo-b

Webhooks¶

What GitHub events does the webhook handle?¶

push → incremental re-index of changed files
pull_request (opened, synchronize, reopened) → blast radius analysis, comment posted to PR

Does repobrain post comments to GitHub PRs?¶

Yes, when the webhook is configured and GITHUB_TOKEN is set. Comments include the blast radius table and risk score.

Privacy & Security¶

Is the PyPI token I used stored anywhere?¶

No. It was used only to upload the package and is not stored in any file.

Can I restrict repobrain to only certain files?¶

Create a .repobrain-ignore file in your repo root (gitignore syntax):

secrets/
*.pem
.env*

Does repobrain store credentials?¶

No credentials are stored by repobrain. The Anthropic API key is read from the environment at runtime.