Cassetto 🧠

Give your AI assistant structural awareness of any codebase.

The Problem

Your AI coding assistant can see your open files. It cannot see your architecture, call graph, or which functions break if you change something. It guesses — and gets it wrong.

The Solution

Cassetto indexes your codebase into a call graph + vector database + keyword index and serves 18 intelligence tools to your AI via MCP. Three commands, 60 seconds, everything local.

pip install cassetto
ollama pull nomic-embed-text
cassetto index . && cassetto setup

That's it. Your AI now calls Cassetto tools automatically — you just talk normally.

What Changes

Benchmarks

Tested on a React + Django project (52 files, 129 symbols):

Full benchmark report with raw outputs: eval/benchmark_report.html

Setup

Prerequisites

• Python 3.11+
• Ollama running locally
• Git installed

Install & Index

pip install cassetto
ollama pull nomic-embed-text

cd /path/to/your/project
cassetto index .

This parses all source files, generates embeddings, builds call/import graphs, analyzes git history, and stores everything locally in ~/.cassetto/.

Connect to Your AI

cassetto setup

Auto-configures MCP for Antigravity (Gemini/Claude), Claude Desktop, and Cursor. Restart your assistant and you're done.

All 18 Tools

Your AI calls these automatically. You never need to know they exist.

Search

Symbol Intelligence

Graph Analysis

Git Intelligence

Architecture

How It Works

Your Code → tree-sitter AST → Chunks (functions, classes)
                                  ↓
                    ┌─────────────┼─────────────┐
                    ↓             ↓             ↓
              LanceDB        SQLite FTS5     DuckDB
            (vectors)        (keywords)    (call graph)
                    ↓             ↓             ↓
                    └─────────────┼─────────────┘
                                  ↓
                        MCP Server (18 tools)
                                  ↓
                        Your AI Assistant

1. AST Parsing (tree-sitter) — Extracts functions, classes, methods as structured chunks across 13 languages
2. Embeddings (Ollama) — 768-dim semantic vectors for each chunk
3. Hybrid Search (LanceDB + SQLite) — Vector similarity + BM25 keyword matching, merged with Reciprocal Rank Fusion
4. Call Graph (DuckDB) — Who-calls-what, component renders, class inheritance, with PageRank scoring
5. Import Graph — Module dependencies, resolved to actual file paths
6. Git Intelligence — Churn rates, ownership, change coupling from git history
7. MCP Server (FastMCP) — Exposes everything as tools via JSON-RPC over stdio

Everything runs locally. No cloud, no API keys required.

CLI Reference

cassetto index [dir]           # Index a project (default: current dir)
cassetto index . --force       # Force full re-index
cassetto setup                 # Auto-configure MCP for your AI assistant
cassetto setup -p myproject    # Configure with specific project ID
cassetto serve                 # Start MCP server manually
cassetto search "auth flow"    # Quick search from terminal
cassetto watch .               # Watch for changes, re-index live

Configuration

Supported Languages

Python · JavaScript · TypeScript · JSX · TSX · Go · Rust · Java · Ruby · PHP · C# · C · C++

Contributing

We welcome contributions! See CONTRIBUTING.md for setup instructions and Good First Issues if you're looking for a place to start.

License

Apache 2.0