llms.txt

The documentation site publishes a machine-readable index at /llms.txt so an agent can enumerate every published page and its summary without parsing the site's HTML.

What llms.txt Contains

The file is a single Markdown document: a top-level heading, a one-line description, an Agent surface section, and then one section per page type. Under each type is a list of its pages, each a Markdown link to the page's exported envelope with the page's summary:

# Docs

Documentation for Dots — every page is a row in the Docs Dotabase.

## Agent surface

The Dots API serves this corpus to agents directly, alongside every other published Dotabase:

- MCP: `POST https://app.mydots.ai/api/v1/public/mcp` — the Model Context Protocol endpoint for the public plane.
- CLI: `dots public search | graph | rows | get | manifest` — the same plane from a terminal.
- REST: `https://app.mydots.ai/api/v1/public/publications/{slug}/...` — the public read plane both of the above are built on.

## start

- [What Is Dots?](/envelopes/what-is-dots.md): Dots is a knowledge platform...

## concepts

- [How Dotabases Work](/envelopes/dotabases.md): A Dotabase is the platform's foundational...

Sections appear in a fixed type order — start, concepts, reference, how-to, tutorials — and a type with no published pages is omitted rather than emitted empty. Within a section, pages are sorted by slug, and a page whose summary is empty renders as the link alone with no trailing colon.

Because the docs corpus is a Dotabase, glossary terms are pages too. They are listed under whichever type each page declares, not under a section derived from its kind — a term written as an onboarding note sits under start alongside the ordinary pages.

The Agent Surface Section

The Agent surface section is the same for every build — it names the three ways an agent can reach the published corpus without going through the site at all. The MCP endpoint speaks the Model Context Protocol; the dots public command family reaches the same plane from a terminal; and both are built on the anonymous REST namespace under /api/v1/public/publications/, which serves any published Dotabase, not just the docs.

The Rest Of The Export

llms.txt is one of several artifacts the site builds from the Docs Dotabase:

• /envelopes/<slug>.md — each page as its row envelope (the same format described in Row Envelope Format).
• /pages/<slug>.json — the same page as structured data: title, id, properties, body, its incoming and outgoing edges, and basic Open Graph metadata.
• /corpus.json — the whole graph in one file: every node, every typed edge, and the counts (total pages, per type, per kind, and edge count), plus the exact commit the export was built from.

Every one of these is derived from the same Docs Dotabase the human-facing site reads, so the numbers, the graph, and the prose never drift apart.