Local markdown mirror of docs.typesafe.ai, crawled with grub-crawler on 2026-09-17. Starting page: Intent routing.
Official index: llms.txt.
Lume lives in lume/. It builds a BM25 index plus a semantic knowledge graph from dict/typesafe.csv, then search walks that graph (intent routing → handler, confidence-gated routing, fan-out, …).
Run these from the typesafe-arena directory. Verified on 2026-09-17: 46 files, 664 sections, graph search for intent routing ranked docs/patterns/intent-routing.md first.
Needs Rust/Cargo (1.75+).
cd lume
cargo build --release
cd ..Binary: lume\target\release\lume.exe (Windows) or lume/target/release/lume (Unix).
.\lume\target\release\lume.exe index --db .lume-index --tag-dict dict\typesafe.csv docsUnix:
./lume/target/release/lume index --db .lume-index --tag-dict dict/typesafe.csv docsThat writes .lume-index/ (bm25.json, entity_graph.json, spelling.json, state.json). The tag dictionary is dict/typesafe.csv — phrases like intent routing, Choice, Noul, confidence become graph nodes.
Re-index after doc changes:
.\lume\target\release\lume.exe index update --db .lume-indexForce a full rebuild:
.\lume\target\release\lume.exe index -f --db .lume-index --tag-dict dict\typesafe.csv docs.\lume\target\release\lume.exe search --db .lume-index "intent routing"Graph boost is on by default (-g 0.4). Disable it with -g 0. More hits: -l 20.
Optional extras (not required for lexical + graph search):
-sdense vectors (needs a NUTS token and Shivvr)-oLLM entity extraction via Ollama
- Cookbooks
- Structure recovery
- Autoresearch feature discovery
- Double-checking citations
- Classification using confidence
- Classifying RAG passages
- Self-consistency: choices
- Self-consistency: nouls
- Date extraction
- Knowledge graph entity alignment
- Function calling
- Hierarchical classification
- Guardrails for LLMs
- Parallel questions
- Pre-parsed value extraction
- Re-ranking
- SDE cascade
- Line-by-line search
- Skill suggestion