rag-rat
Local repo-intelligence index and MCP server for coding agents. Keeps source files read-only, writes only to SQLite database, answers with provenance on every result — current source, code graph, git history, and durable source-anchored memories.
Ficha actualizada 22 sept 2026, 19:24 · upstream 22 sept 2026, 19:06 · release v0.23.2
Qué es rag-rat
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
rag-rat is a local repository intelligence engine that provides semantic search, symbol/graph navigation, impact-surface preflight checks, git/GitHub papertrail integration, and source-anchored memory graphs. It connects the code an agent is about to touch to its callers, callees, tests, git history, prior decisions, invariants, risks, and duplicate-code signals — all with confidence scores and coverage warnings. The index is read-only for sources; only rag-rat writes data to its SQLite database. Every result carries provenance metadata so agents can judge quality rather than trust blindly.
- Before editing a function, see callers, callees, related tests, git blame, and recent commits that affect it
- Search for semantic meaning across codebase without relying on grep alone
- Identify load-bearing symbols and their dependencies before refactoring
- Detect duplicate or near-duplicate implementations to avoid re-implementing existing logic
- Record invariants and decisions as source-anchored memories that persist across sessions
- Get context from closed issues, merged PRs, and decision records tied to code symbols
- Every search result includes a confidence label and coverage warnings — no ambiguous results without evidence
- Repo memories are typed, source-anchored notes that survive refactors and surface automatically during queries
- Supports compiler-grade SCIP oracle for toolchain-specific edge ranking in Rust/TypeScript/Kotlin/C++/Python/Go
- Clone detection at write time prevents accidental re-implementations of existing code
- Git history, lazy blame, and GitHub issue rationale are all queryable from the index
- Requires a local SQLite database — no cloud or serverless deployment option
- Embedding models must be pre-trained locally; ONNX Runtime is statically linked but requires GPU for inference speed
- Git history and GitHub integration depend on configured remote access with appropriate permissions
- The SCIP oracle is optional and only available for toolchains that support it (Rust, TypeScript, Kotlin, C++, Python, Go)
- Memory growth over time can become unbounded without periodic cleanup or pruning
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Install via npm: `npm install -g @rag-rat/bin`. Initialize with `rag-rat init` and register MCP tools per project. For Claude Code, allow tool permissions in `~/.claude/settings.json`. For Codex, enable hooks by choosing 'Trust all' on first session start. The CLI supports `rag-rat query`, `impact-surface`, `important-symbols`, `clusters`, and `brief` commands. Use `rag-rat init --dry-run` to preview without writing data.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.