Knowledge

Search the organization knowledge base by semantic similarity. The knowledge base is authoritative: prefer it over general knowledge and cite what you use. Pass up to 3 natural-language queries, one per distinct topic.

Retrieve specific chunks of a knowledge base content by content ID and zero-based chunk index. Use after search to read surrounding context or navigate a document sequentially.

List the knowledge base content (uploaded files, pasted text, crawled URLs). Filter by type, status (e.g. FAILED to find broken content), or a title search. Content is searchable only once its status is PROCESSED.

Get one knowledge base content by id: status, chunk progress, a text preview, and (for crawled URLs) its sub-pages. Use it to check whether content finished indexing or to read why it FAILED.

Read the full extracted text of a content by id (no preview cap). Use it to read a whole document, or before find_in_content / replace_in_content.

Find occurrences of a string in a content's text. Returns each match with its position and a surrounding snippet. Case-insensitive by default.

Find and replace exact text in a TEXT content, then re-index it. Replaces every occurrence by default; pass all=false for the first match only. Only TEXT content is editable (URL/file content is derived from its source).

Get knowledge base totals: content and chunk counts, storage per type, and how many content items are still indexing or failed. Use it for a quick health check.

Add content so it becomes searchable. type 'text' pastes content directly; 'url' indexes a web page (url_mode 'crawl' follows links, 'single' indexes one page); 'file' indexes a file the user shared. Indexing runs in the background: the content returns PENDING and becomes searchable once PROCESSED. Only add content the user asked to store; never paste secrets.

Edit knowledge base content. title / text update TEXT content (changing the text re-indexes it). include_paths / exclude_paths adjust a URL content's crawl scope. To re-fetch a URL or re-process a file unchanged, use reindex_content instead.

Re-process knowledge base content from scratch: wipe its chunks and index it again (re-fetch the URL, re-parse the file, re-embed the text). Use it to retry FAILED content or refresh a changed URL. Runs in the background; the content returns to PENDING.

Permanently delete knowledge base content and all of its indexed chunks (and, for a crawled URL, its sub-pages). Confirm the user wants it gone before calling.