docs-reading

Overview

docs-reading is a two-phase pipeline:

1. Parse — A human runs a parser script once to ingest documentation into a local SQLite database with FTS5 full-text search and optional sentence-embedding vectors. Multiple corpora and versions can coexist in a single DB.
2. Query — Claude calls agent-docs-search (a self-contained binary on $PATH) to search, browse, and retrieve documentation without loading the raw files into context.

Parsing (human setup step)

Generic Markdown

uv run scripts/parse_docs.py \
  --input /path/to/docs \
  --db ./docs.db \
  --corpus-name mylib \
  --corpus-version 1.0