Add knowledge to a Mind

Attach new knowledge to a Mind. Pick the input by content type and body shape: (1) multipart/form-data with a file part uploads a document or image (max 50 MB); (2) JSON { file: { name, url, type? }, description } ingests a file the server fetches itself from a public URL or a Minds workspace upload URL (SSRF-protected, 20 s fetch timeout, same 50 MB limit); (3) JSON { link, description } ingests a web page, YouTube video, or paper URL; (4) JSON { keywords: [...] } runs a web search per keyword and adds what it finds. JSON bodies are dispatched in the order file, then keywords, then link, so send exactly one shape. Modes 1-3 create one knowledge item, return it with processing.status: queued and a statusUrl. For files, the text is indexed before the response when it can be read directly (searchable: true), so the Mind can use the file at once; tables and charts are added in the background. Otherwise poll that URL (or GET .../knowledge/{itemId}/status) until readyForRetrieval is true before relying on the content, especially with sourcePolicy: knowledge_only. Mode 4 queues a Mind-level enrichment job instead (no item is returned; keywords are trimmed, de-duplicated, and capped at 15); track it via GET /v1/minds/{mindId}/training. regeneratePrompt (rebuild the system prompt after ingestion succeeds) defaults to false for files and links and true for keywords. Access: you must be able to edit the Mind (owner, direct collaborator, or member of the team the Mind is shared with); other Minds answer 404 Mind not found or access denied.