roteiro

A provenance-tagged knowledge graph for your codebase — structure, intent, and context in one queryable store, for humans and AI agents alike.

Roteiro reads a git repository and assembles a single SQLite knowledge graph of the code (symbols, calls, imports across 15+ languages), the documents and decisions that govern it (ADRs, blueprints, // @rto: annotations), and the fuzzy relationships between them (embeddings). Every node and edge records how it was produced — derived, authored, or inferred — so you always know whether a fact is a verified truth, a human decision, or a machine's suggestion. Offline by default; git-native and content-addressed, so the graph is shareable across a team.

This crate is the umbrella CLI. Common commands:

roteiro init      # scaffold the store + git hooks + AGENTS.md in your repo
roteiro sync      # build / incrementally update the graph
roteiro query …   # explain a node and its provenance-labelled edges
roteiro review    # graph-grounded review of your current change
roteiro check     # verify authored intent against the code (a drift gate)
roteiro render …  # emit a docs site or an Obsidian vault

Feature flags gate the heavier capabilities (inference, inference-local-models, pdf-text, image-ocr, image-vision, serve, mcp, exec-boxlite); the default build is small, needs no C++/cmake/libclang, and makes no network call on its own. It carries models and exec-subprocess, so the whole of "prepare once, then work offline" — roteiro model pull and roteiro security prefetch|status|run — exists without a rebuild. Presence is not activity: pull fetches nothing until you consent, and security run refuses without --allow-unsandboxed every time. See cargo install roteiro --all-features for everything, and --no-default-features --features execution for a build that provisions and ingests but cannot execute an analyzer.

• Docs & guide: https://roteiro.dev
• Source & issues: https://github.com/OffeneDatenmodellierung/Roteiro

Dual-licensed under MIT or Apache-2.0.