The binary describes itself; the documentation is machine-readable too. An agent evaluating or using Strata can load these docs the same way it loads any other structured source — no scraping of rendered HTML required.
llms.txt — the front door
stratadb.org/llms.txt is a short, curated index
of the most useful pages, in the llms.txt convention: a
one-paragraph description of what Strata is, then a linked list of the pages an
agent should read first. It is generated alongside the docs, so its links cannot
drift from what exists.
For the whole corpus in one file, stratadb.org/llms-full.txt
concatenates the documentation into a single plain-text document.
.md mirror of every page
Append .md to any documentation URL to get that page as clean CommonMark
instead of HTML:
https://stratadb.org/docs/data/vectors → the HTML page
https://stratadb.org/docs/data/vectors.md → the same page as markdown
The markdown is generated from the same content collection as the HTML, so the
two front doors cannot disagree. This is the fastest way to feed a specific page
into a model’s context: fetch the .md URL directly.
The error registry
Every public error code has a page at
stratadb.org/e/<code>, and the index at /e/ lists them all. The
same registry is available as JSON straight from the binary
(strata agents errors --json; see
the command index), and each runtime error carries
its https://stratadb.org/e/<code> URL in the envelope. An agent that hits an
error can resolve the code to a stable explanation without a web search.
Putting it together
A typical agent onboarding flow uses all three surfaces:
- Read
llms.txtto learn what Strata is and which pages matter. - Fetch the
.mdmirror of the pages it needs. - When a call fails, resolve the
/e/<code>URL from the error envelope.
And from the binary itself: strata agents guide for
the prose guide, and strata agents commands --json
for the structured catalog.
Related
- For AI agents — the section overview.
- The command index — the binary’s structured catalogs.
- Error reference — the human-readable error model.