Integration API docs split per endpoint for sharper agents

Until March 25, when the workflow builder or the chat agent needed a service, it loaded that service's complete API reference into its context, however few endpoints the task used. A large reference crowded out the conversation and made mistakes more likely. We split every integration's documentation into individual operations, one per endpoint, plus one shared section per service covering authentication, pagination, and error handling.

How it works

• Search, then fetch. Agents search across all operations with hybrid retrieval (vector embeddings plus full-text) and get the top 10 matches grouped by service, then fetch only the operations they need.
• Everything needed to call it. Each fetched operation arrives with its service's shared section and the exact credential fields its code should read.
• No more whole documents. The path that loaded an entire reference was removed, so context holds only the operations an agent looked up.
• Shorter prompts. Moving reference material behind search tools also shortened the agents' standing instructions, which leaves more room for your conversation and your data.

Browse every documented integration in the integrations catalog.