Index an OpenAPI spec so the assistant can answer questions such as “How do I create a webhook?” or “What fields does the invoice object have?”, with citations into the spec. Available on every plan.

​
Before you start

  • The spec at a public URL, for example https://api.example.com/openapi.json. Kelu fetches it without credentials.
  • OpenAPI 3.x or Swagger 2.0, in JSON or YAML, up to 16 MB.

​
Set up

1

Add the source

Open the knowledge base, go to Sources → Add source, pick OpenAPI Spec and click Continue. Enter a Display name.
2

Enter the spec URL

Paste the Spec URL. If the spec is on GitHub, use the Raw file URL (https://raw.githubusercontent.com/...). Click Test connection, then Add source.
3

Ask an API question

In the knowledge base’s Playground tab, ask about an endpoint or a schema field.

​
What gets indexed

  • One document per endpoint, with its method and path, summary, description, base URL, auth, parameters, request body, responses and examples (including x-codeSamples). $refs are resolved.
  • One document per tag, listing the endpoints under it. Endpoints with no tag go under default.
  • One authentication document, from the spec’s security schemes.
  • One document per data model, with its fields, types, enums, defaults and required flags, plus one document that lists every model.
Citations link to the spec URL with a fragment such as #operation/createWebhook, #tag/Billing, #authentication or #schema/Invoice.

​
Sync

The spec is fetched again every hour. Sync Now on the source page runs a sync at once — useful right after you publish a new version.

​
Troubleshooting

  • “HTTP 4xx fetching spec” — the URL needs a login. Publish the spec where anyone can fetch it.
  • “not valid JSON or YAML” or “not an OpenAPI specification” — the URL returns something else, such as an HTML page. Point it at the raw spec file.
  • “declares no paths” — the spec has no endpoints to index.
  • Every endpoint is under default — add tags to the operations in your spec.

​
Works well with

  • A website source for your API guides, so the guides and the reference answer together.
  • The MCP server, so developers can query your API reference from their editor.