Local code intelligence for agents and developers. GraphIQ indexes a repository into a structural graph of symbols, calls, imports, constants, and type flow — then searches it with ranked retrieval that understands how code is connected, not just what strings it contains. No embeddings, no LLM, no network calls. Everything lives in a single SQLite file.
6,461,740 lines · 433,898 symbols · 726,439 edges · 5 languages. The largest codebase ever benchmarked with GraphIQ (~20× prior scale). 100 questions per metric. Full report · query sets.
| Metric | GraphIQ | Grep | Δ |
|---|---|---|---|
| NDCG@10 | 0.201 | 0.123 | +63% |
| MRR@10 | 0.558 | 0.343 | +63% |
| Hit@10 | 0.85 | 0.52 | +33 pts |
| Categories won | 8 / 10 | 2 / 10 | — |
- Install — Homebrew, install script, or build from source
- Quickstart — index, search, wire into a harness
- How it works — the indexing + search pipeline
- What gets indexed — languages, symbol types, graph edges
- CLI reference — every command and flag
- MCP tools — the tool surface agents see
- Supported harnesses — Claude Code, Codex, OpenCode, Cursor, Pi, and more
- Desktop app — the visual index browser
- Reliability & resource control — CPU caps, dormancy, crash-safety
- Performance & benchmarks — measured quality vs grep
- Benchmarks — full methodology and per-codebase results
- Signet plugin — the managed plugin for Signet users
- Research notes — 29 phases of experimentation
- Hardening roadmap — reliability work, past and planned
- Contributing — building, testing, and CI
graphiq index /path/to/project
graphiq search "rate limit middleware"See Quickstart for the full flow, including wiring GraphIQ into your AI coding agent.
MIT — see LICENSE.