For coding agents
Everything here is a flat file on a CDN. There is no auth, no rate limit and no query planner between you and the corpus, so a read costs one GET and an ETag makes a re-read free.
Start here
| Path | What it is | Size |
|---|---|---|
/llms.txt | The map: every section with a one-line gloss. | small |
/llms-full.txt | Curated full text, capped so it fits a context window. | large |
/docs/index.json | The catalogue: 257 entries, no bodies. Hold this, fetch precisely. | small |
/docs/<slug>.md | One page as markdown. Half the tokens of HTML. | small |
/docs/<slug>.json | One page structured, with cross-links and provenance. | small |
/docs/graph.json | Composes and used-by across the ecosystem. | small |
/docs/symbols.json | Which page documents an exported name. | small |
/docs/search-index.json | Prebuilt search. Rank locally, no round trip. | medium |
/docs/downloads/index.json | Installable agent skills. | small |
Or connect over MCP
The leumas-docs MCP server exposes searchDocs, getPage,listPackages, getSymbol and howDoI. Every result carries the file and commit it came from, so you can cite rather than paraphrase, and a stale corpus is detectable from the outside.
Citing
Each page carries source.path and source.commit. Quote them. If the commit is older than the file you are reading in a checkout, this corpus is behind and you should say so rather than answering from it.