Project
hermes-kb
200 pages of Hermes Agent docs in one SQLite file. Searchable offline.
What this is
A local, offline knowledge base for the Hermes Agent documentation: about 200 pages crawled once, stored in a single SQLite database with FTS5 full-text search. Querying it needs no dependencies and no network. One file, instant answers.
Why I built it
Agent documentation lives on the web, which is exactly where you don’t want a local agent reaching every time it forgets a setup detail. Crawl once, search forever. The database travels with the bot.
How it works
A build script crawls the docs and writes them into SQLite with an FTS5 index. A small query interface searches it from the command line or from code, and a format_for_llm() helper returns results shaped for dropping straight into a model’s context. Rebuilding after a docs update is one script; only the rebuild step needs network access.
What it can do
- Full-text search across the entire documentation set, offline
- Zero dependencies at query time; the
.dbfile is the whole distribution - One function call to feed clean, formatted results to any bot or script
Limitations
It knows exactly what was crawled and nothing newer; freshness is a manual rebuild away. Search is lexical FTS5, not embeddings, which turns out to be plenty for documentation lookups and keeps the whole thing dependency-free.
Status and next
Stable and in regular use. It feeds a private agent setup I run at home; that part is not published.