Tools

trusty-search — documentation

Machine-wide hybrid code-search service: BM25 + vector + knowledge-graph search behind one always-on daemon and an MCP server. Crate lives in crates/trusty-search/.

This directory is the extended index for trusty-search research, regression testing, historical design, and engineering-session documentation. The crate README and rustdoc are the current package entry points.

Documentation map

Subdir	What's here
`spec/`	Historical PRD, architecture, and component baseline, indexed in `spec/README.md`.
`research/`	Investigation, audit, and decision documents — BM25 memory, Candle/Metal validation, the nested-index fan-out RFC, NLP/ER/KG indexing, the staged-pipeline (stage-1 minimal, stage-3 KG, phase-3 async symbol-graph) decisions, and the trusty-search vs. mcp-vector-search comparison. Indexed in `research/README.md`.
`regression-testing/`	Versioned performance snapshots (`v{VERSION}-{DATE}.md`) plus alternate-corpus baselines (synthetic, open-mpm) and certification runs. `current.md` symlinks the latest snapshot. Methodology in `regression-testing/README.md`.
`decisions/`	Crate-specific Architecture Decision Records (Nygard format). Indexed in `decisions/README.md`.
`sessions/`	Engineering-session narratives (`SESSION-{DATE}-{topic}.md`). Indexed in `sessions/README.md`.
`examples/`	Reference configurations: `trusty-search.yaml` — multi-index per-repo config consumed by `trusty-search index`.

Where to start

What is trusty-search? Start with the crate README; use the behavior-contract catalog for current target-state specs.
Performance / benchmarks? regression-testing/README.md → regression-testing/current.md.
Why a feature works the way it does? research/README.md.
Configuring multi-index repos? examples/trusty-search.yaml.

Conventions

Subdirs follow the workspace documentation conventions described in the root CLAUDE.md. research/ files are dated point-in-time investigations preserved as-is; regression-testing/ snapshots are tied to released versions.