An AI that knows your whole API — endpoint by endpoint.
Stop answering the same questions. A floating Ask AI button on every public docs site opens a chat trained on your spec. Visitors get straight answers with real code examples — and every endpoint citation is a clickable pill that jumps the docs to that exact endpoint.
/v1/usersUse the page and per_page query params on listUsers. Max 100 per page.
&per_page=50'
The response includes a meta.pagination object with total_pages.
< 800ms
First-token latency
50+
Endpoints in inline context
200k
Tokens / day on Business
100%
Grounded — citations only
Grounded retrieval
Three stages, then an answer with citations.
Up to ~50 endpoints, the entire spec is passed inline. Beyond that, a pgvector retriever pulls only the operations the question needs. Either way, the model only sees your spec — and is told to admit when something isn't documented rather than guess.
- 1
Visitor asks
“How do I paginate /v1/users?” The drawer streams the question into the chat panel and starts a conversation in the visitor's localStorage session.
- 2
Retriever scopes
For specs ≤50 endpoints the whole document goes inline. Larger specs go through a pgvector top-k retriever scoped to the project's spec embeddings.
- 3
Answer + citations
The model emits structured citation tokens for every endpoint it references. The frontend renders them as clickable pills that scroll the docs page directly to that endpoint.
Deep-link citations
Every reference is a clickable pill.
The model emits structured citation tokens for every endpoint it references. The frontend renders them as small rounded pills using the project's accent colour, and clicking one scrolls the docs body to that operation — visitors never leave context.
- Visitor never leaves the docs. The pill scrolls in-page, not to a new tab.
- Citations match operationId. Re-rendering the spec moves them; the model gets fresh identifiers on every page load.
- Code blocks copy with one click. cURL / Fetch / Python all on the same chat-message bar.
Use the listUsers endpoint. Pass page and per_page.
/v1/usersList users
Returns a paginated list. page and per_page are query parameters; max 100 per page.
Conversation analytics
The questions visitors actually ask.
Every conversation lands in the project's Chat Activity tab — the highest-signal feedback your docs will ever get. It tells you which endpoint descriptions are too vague, which auth flow confuses people, and which migration path still hasn't been documented.
Ticket deflection, measured
See the queries support staff would otherwise be answering. Drop a top-asked question into your docs and the next visitor finds it themselves.
Streaming + persistent
Streamed completions, per-visitor session in localStorage, copy-to-clipboard on every code block. Token-budget readout in the drawer footer.
Per-turn token accounting
Daily caps with an 8-message sliding window. Free 0, Pro 25k, Business 200k. Cap resets at midnight UTC; hitting it shows a polite throttle, not a 500.
Plan tiers
Daily-token budgets you can predict.
Free
0 tokens / day
- AI search (5 queries / month)
- No floating chat drawer
- Upgrade to enable
Pro
25,000 / day
- Floating chat on every docs site
- Streaming completions
- Citations + deep-links
- Conversation analytics
- 8-message sliding window
Business
200,000 / day
- Everything in Pro
- 8× the daily token cap
- Custom domain on the docs site
- Priority support
Pairs well with
MCP Server
Same retrieval engine, exposed as a tool for AI assistants. Ask AI in the docs; same brain answering in Claude / Cursor.
Docs page builder
Customise the surface the chat drawer lives on — logo, header, per-endpoint markdown — without touching MDX.
SDK Generator
When the chat answers “how do I list users?”, the SDK is the next click — typed methods drop straight into the visitor's repo.
The chat that knows your spec.
Pro plan flips the floating Ask AI drawer on every public docs site. 25k tokens/day at $9/month, 200k on Business.