This repository was archived by the owner on Oct 8, 2026. It is now read-only.
Repository navigation
v0.7.0
First public release of Tempograph.
What's in the box
- Graph engine: tree-sitter parsing across 170+ languages, SQLite graph with WAL mode, content-hashed caching (branch switching doesn't trigger rebuilds)
- 24 MCP tools:
prepare_contextis the main one. One call gives your AI agent the right files for any task. Also includes focus, blast radius, hotspots, dead code, diff context, and more - Hybrid search: FTS5 keyword search + sqlite-vec embeddings + structural graph, combined via Reciprocal Rank Fusion
- Validated results: +18.6% F1 on change-localization (p=0.049, n=45). No other code context tool publishes retrieval benchmarks with statistical significance
- Fast: ~21ms warm queries. Only re-parses changed files on incremental builds
- CLI + MCP + Python API: use it however you want
- Kits: composable multi-mode workflows (5 built-in, custom kit support)
- Docker support: run the MCP server in a container
- 4700+ tests
Install
pip install tempographOr see the README for quickstart script, Docker, and optional extras.
License
BSL 1.1. Free to use, read, modify, and fork. Can't be used as a commercial hosted service. Converts to Apache 2.0 in 2030.