MAP Documentation

MAP docs publish two machine-readable artifacts for agent consumption.

`/llms.txt`

A compact index. Each line is a doc URL with a one-line description. Intended as a low-token directory to drop into agent context.

# MAP Documentation Directory
# https://docs.multiagentic.dev

/docs/quickstart        : From install to first dispatched call.
/docs/concepts/overview : What MAP is and isn't.
/docs/concepts/estates  : Protocol, Agent, Hybrid, Adapter.
...
/docs/protocols/macs    : MACS — single port of entry. Verifies signatures, derives capabilities.
/docs/protocols/mind    : MIND — cognitive substrate. Four memory types, HNSW search.
...

Fetch:

curl https://docs.multiagentic.dev/llms.txt

`/llms-full.txt`

The full corpus flattened to plain text. Every doc page concatenated in a stable order. Intended for total context drops (very large) or for offline indexing.

curl https://docs.multiagentic.dev/llms-full.txt > map-docs.txt
wc -l map-docs.txt
# ~25k lines for the full corpus

The file is regenerated on every deploy. Use the ETag header for cache-aware refreshes.

Generated, not authored

Both files are generated from the MDX content at deploy time. The generator walks content/docs/, reads each MDX file's frontmatter and body, and emits:

llms.txt — index with frontmatter title + description
llms-full.txt — full body (after MDX → plain-text conversion)

If you change a doc page, the generated files update on the next deploy.

Usage patterns

Pattern A: directory in context

System prompt:
"You have access to MAP via MCP. For docs, read /llms.txt for the directory,
then fetch specific pages with curl when needed."

[Agent fetches /llms.txt once at session start; only pulls full pages on demand.]

Pattern B: full corpus pinned

System prompt:
"The following is the complete MAP documentation. Reference it directly.
Do not invent operations that do not appear here."

[/llms-full.txt pasted into the system prompt — ~150k tokens. Works for
long-context models. Total coverage, zero retrieval errors.]

Pattern C: retrieval over the corpus

[Embed /llms-full.txt offline. Index with FAISS / Qdrant / pgvector.
At query time, embed the question, retrieve top-k chunks, pass to LLM.]

This is what we recommend for production agent fleets. The corpus is small enough to embed cheaply; retrieval gives you the right page without flooding context.

Live search API

Programmatic search over the docs without needing to host your own retrieval:

curl 'https://docs.multiagentic.dev/api/search?query=audit+chain'

Response:

{
  "results": [
    {
      "url":   "/docs/concepts/audit",
      "title": "Audit & explainability",
      "score": 0.94,
      "snippet": "Every refusal, every grant, every dispatched operation enters the audit chain. The chain is hash-linked..."
    },
    ...
  ]
}

The search index is built from the same MDX corpus. Backed by Orama on the server side.

We don't enforce auth on /llms.txt, /llms-full.txt, or /api/search. The docs are public. Your API calls require an API key; reading the docs does not.

llms.txt