Blink Query

blink-query

**A typed wiki for LLMs — markdown on disk, resolution in the library.**

[](https://www.npmjs.com/package/blink-query) [](LICENSE) [](#development)


blink-query is a knowledge resolution library for LLM agents. Markdown files stay on disk where you can read and edit them; the library adds typed records, title-weighted BM25 search, deterministic path resolution, and an MCP server — all in a SQLite file next to your notes. No cloud, no embeddings, no daemon.

Features

• 5 typed record types — SUMMARY, META, SOURCE, ALIAS, COLLECTION — each carries a consumption instruction for the agent (read the summary / parse JSON / fetch the source / follow the redirect / browse children)
• Namespaces — slash-delimited paths group records by topic (e.g., wiki/concepts, decisions/auth, people/arpit); each record has a stable path at namespace/slug-of-title
• Title-weighted BM25 over FTS5 — title matches ranked 10× over body summary, with per-type rank offsets that promote canonical pages
• Hub-vs-leaf default classifier — automatically promotes index / README / Home files in hub directories to SUMMARY, no per-corpus tuning
• Path resolution — deterministic O(1) lookup by namespace/slug, with NXDOMAIN / STALE / ALIAS_LOOP statuses
• MCP server with 11 tools — drop-in for Claude Desktop, Claude Code, Cursor, Codex
• Query DSL — Peggy-parsed: wiki where type = "SUMMARY" and tags contains "auth" limit 10
• Multi-source ingestion — markdown directories, PostgreSQL rows, GitHub Issues, git repos, web URLs
• Zones — top-level namespaces with default TTL and required-tag policies
• [[wikilink]] extraction — auto-creates ALIAS records on ingest

---