B.A.T.H.O

Give your AI coding agent a map of your codebase — not the whole territory.
Reduce token spend 10x, eliminate hallucinations, and ship faster with graph-powered code intelligence.

Works with: Claude Code · Cursor · Windsurf · Antigravity · Gemini CLI · Cline · OpenCode · Aider

Quick Setup

The fastest way to set up Batho: give the skill file to your AI agent and let it do everything for you.

1. Download the skill file

curl -O https://raw.githubusercontent.com/sageoz/batho/main/SKILL.md

2. Give it to your AI agent

Paste this into your agent's chat (Claude Code, Cursor, Windsurf, or any agent that supports skills):

Read SKILL.md and set up Batho for this repo

3. Your agent handles the rest

• Installs Batho globally (pip / uv / pipx)
• Builds the code graph for your repository
• Auto-detects all installed AI clients (Claude Desktop, Cursor, Windsurf, VS Code)
• Configures MCP for each client
• Verifies the setup works end-to-end

No manual JSON editing. No config file hunting. Your agent does it all.

Multi-repo? Your agent can register multiple repos via the MCP registry — one config serves all. See the Multi-Repo Guide.

# Install
pip install batho

# Build the code graph
batho build --root . --verbose

# Start the MCP server
batho mcp

{
  "mcpServers": {
    "batho": { "command": "batho", "args": ["mcp"] }
  }
}

Why Batho?

AI coding agents are powerful — but they burn tokens reading files and hallucinate when context is thin. Batho gives your agent a structured code graph so it works smarter, not harder.

• Slash token costs — Your agent queries a graph instead of reading entire files. 10x fewer tokens per task — no more dumping your repo into the LLM.
• Eliminate hallucinations — Deterministic, tree-sitter-parsed relationships. Your agent gets facts, not guesses — zero hallucinations on structural queries.
• Agent superpowers — Bug tracking, security audits, refactoring, code review — your agent handles more, accurately. When cost and quality are solved, automation widens with imagination.

MCP Tools

Batho exposes 10 MCP tools your AI agent can call:

Full reference: docs

Features

• 40+ languages — Python, TypeScript, Rust, Go, Java, C/C++ and more via tree-sitter
• 10x token compression — your agent uses a fraction of the context window
• Zero hallucinations — deterministic AST-parsed relationships, not embeddings or guesses
• Fast incremental updates — hash-based change detection re-parses only modified files
• Cross-file symbol resolution — your agent sees how functions, classes, and dependencies connect
• 38 built-in analysis plugins — security, quality, and optimization rules with custom rule support
• Time-machine — node-level diff history across every indexed run
• Zero code execution — safe to run in CI or on untrusted repositories
• MCP-native — works with 8 AI coding agents out of the box

CI/CD

Batho's CI/CD strategy is incremental: download the previous artifact → batho load → batho patch → batho export → upload.

GitHub Actions composite action:

- uses: sageoz/batho@v1.2.0
  with:
    root: "."
    artifact-name: "batho-index"

Full CI/CD guides (GitHub Actions, GitLab CI, reusable workflows): docs

Configuration

Batho runs with zero config. To customize, copy batho.yaml.example to ./batho.yaml.

Full configuration reference: docs

Developer Setup

git clone https://github.com/sageoz/batho.git
cd batho
uv sync --all-groups --all-extras
uv run pytest
uv run python batho_cli.py --help

Documentation

• Quick Start — CLI setup guide
• Setup with AI Agent Skill — Let your agent set up Batho
• MCP Server — Connect AI agents to your code graph
• Whitepaper — Deep technical reference
• CLI Reference — Complete command documentation
• CI/CD — GitHub Actions, GitLab CI, and more

Acknowledgments

Co-authored with Devin — autonomous AI software engineer by Cognition.

Citation

If you use Batho in your research, please cite it as:

@misc{Sharma_Batho_2026,
  author = {Sharma, Rishiraj},
  doi = {10.5281/zenodo.21407508},
  month = {7},
  title = {Batho: Deterministic Code Intelligence Engine},
  url = {https://pypi.org/project/batho/},
  year = {2026}
}

You can also find the citation metadata in CITATION.cff or use GitHub's "Cite this repository" button.

License

Apache License 2.0 — see LICENSE for details.