Skip to main content
All paths below are relative to /m2m/v1. Send Authorization: Bearer <key>. Tenant identity comes from the key. Use knowledge:read for discovery and status and knowledge:write for ingestion and management. Existing quotas, supported file formats, and activation semantics apply.
Text is limited to 200,000 characters; title to 160; sourceId to 128. Updating a source of a different type is rejected. Text/file/URL ingestion returns 201 on creation or 200 on update. A new source can be saved inactive when quota is exceeded; check active and quotaExceeded. Updating an active source beyond quota returns 409. The current knowledge model is source-based: no knowledgeBaseId is required. File uploads accept optional title and assetUpload; sourceId is not supported for file uploads. Crawl JSON accepts url, crawlDepth (1–3, default 2) for legacy URL crawling; managed websites use 1–10, and maxPages (1–500, default 500).

Extended source, website and asset management

This October 13 local source continuation is additive and requires compatible RAG/API deployment and refreshed MCP tools. It exposes existing product operations; no analytics or extraction pipeline is added. Discover sourceId/websiteId/chunkId/folderId through their corresponding lists/results; they are different identities. Source detail includes current text, activation, scope, asset validation/sharing and website generation metadata. includeInactive=true helps diagnose inactive knowledge. Title/URL query is metadata search; semantic search may call an embedding provider and differs from the guest runtime planner/scanner. mode=fallback_all is a bounded unranked fallback, not a semantic match; totalActive is an eligible count, not returned count. Never perform a fresh provider search merely to explain a saved answer. Managed website creation supports depth 1–10 (default 2) and maxPages 1–500 (default 500). Read crawl status/error, actual selection/current generation and usage after acceptance; failed refresh can retain previous knowledge. Recent job lists are bounded, not complete history. Pages remain subject to quota. Source/item pages default 50/max 200; website/folder pages default 50/max 100. Follow nextCursor with identical filters until absent. assetUpload=true and images use existing asset-only ingestion: no KB chunks/token use. ValidationStatus=ready, sharing activation, property/folder scope and folder activation govern guest attachments. KB activation cannot convert asset-only files into knowledge. Files must be smaller than 10 MiB; there is no in-place file replacement/sourceId input. Text edits replace the complete discovered text source. Folder deactivation/deletion retains link revocation; deletion removes member assets. Signed original URLs are private and expire. Reconcile uncertain mutations through discovery/status reads before retrying. For current tone/instructions/escalation use AI settings. For historical answers, saved traces/RAG receipts can provide captured evidence; current knowledge/settings/generations do not prove historical configuration, relevance or causality. Disclose capture/version/linkage gaps.