Most knowledge-base servers stop at read and write. Here an agent can search what customers actually asked, read a conversation with its feedback, then generate an article suggestion and write the document — the whole cycle from complaint to published answer, without anyone exporting a CSV. The citation-scoped answer tool is the other careful piece: answers can be restricted to particular collections, so internal notes do not end up quoted to a customer.
Brainfish's endpoint over their support knowledge base. Twenty-four tools in four groups: documents, collections and catalogs, AI answering, and the conversation history from their chat widget.
- Documents: semantic search, fetch by ID or slug, list, create, update, move between collections, soft or permanent delete
- Collections and catalogs — create, update, delete, and full content sync into a catalog
- brainfish_generate_user_answer — an AI answer with source citations, optionally scoped to particular collections
- brainfish_generate_article_suggestion — asynchronous suggestions for articles the knowledge base is missing
- Sessions: search conversations, read the full detail with feedback and analytics, or the chronological event timeline
OAuth 2.1 with the full discovery set, or a static bearer token beginning bf_api_. The server is documented as stateless; data residency and customer-managed keys are enterprise options.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen