Your customers now ask AI tools before they open your docs. The Agent Layer publishes your documentation in the formats those tools read best, with the evidence to back every answer, so nothing gets guessed.
llms.txt · MCP endpoint · structured JSON · source evidence
Trusted by the AI tools your readers already use
What's the default request timeout?
search_docs called from inside the editor · 2m ago
MCP endpoint connected
4 tools, read-only: no scraping, no stale copies.
/guides/long-running-jobs
evidence: bono@a1f9c2 · rank 2 of 3
/sdk/configuration
evidence: bono@a1f9c2 · rank 1 of 3
60 seconds before aborting
Grounded, never guessed · 2 sources
[1] /sdk/configuration · a1f9c2
[2] /guides/long-running-jobs
Freshness checked · 4m ago
Everything an AI tool needs to answer accurately about your product, served from the same source your customers read.
1# Facts first, layout gone
6124 more pages
A compact index that fits an agent's context budget
Docs an agent can
actually read.
https://docs.yourproduct.com/api/mcp
search_docs(query, limit)
query your docs live, ranked by relevance
read_page(pageId)
pull the current page, cited by commit
list_pages()
the whole site map, returned as data
agent_readiness()
score how well the docs serve agents
Standard Model Context Protocol, inside the editor
Tools, not scraping.
search and read built in.
Structured JSON
Concept references and blocks as data, so pipelines and custom agents parse meaning, not markup.
Evidence links
Every fact traces to a commit and symbol in your product: the receipts behind each answer.
Freshness signals
Each surface carries when it last synced with the product, so agents can prefer current facts.
Scoped access
Decide which surfaces are public to agents and which stay internal: granular, revocable, read-only.
An agent guessing about your API is worse than no agent at all. The Agent Layer replaces plausible-sounding fiction with sourced fact.
Confidently wrong
- Answers from training data months or years out of date.
- Invents parameters and defaults that never existed.
- No way to check where an answer came from.
- Sounds equally sure whether it's right or hallucinating.
Sourced and current
- Reads your live docs, regenerated on every product change.
- Cites the exact page and commit behind each claim.
- Says less when it knows less. No evidence, no assertion.
- You control what's exposed and can revoke it anytime.
This is your documentation answering through the Agent Layer. Pick a question, then watch it retrieve from the graph, answer in plain terms, and show exactly where each fact came from.
Turn on the Agent Layer and your documentation becomes an MCP endpoint and llms.txt feed: current, cited, and scoped exactly how you want.