Knowledge with Guaardvark (MCP tools)

Everything here is the guaardvark MCP server. All retrieval is local (Ollama embeddings, hybrid search, cross-encoder reranking on the user's GPU).

Find and read

tool	use
search_knowledge_base	the question in plain words; returns passages with sources. Cite them.
summarize_corpus	what the whole knowledge base covers, before searching blind
list_documents	which documents are indexed and how many passages each has
get_document_outline	sections or pages of one document, in order
read_document_section	the actual text of one section or page, no search
process_file	extract content from a PDF, DOCX, CSV, Excel, image or other file the user names
fetch_url	title, description and main text of one web page
web_search	DuckDuckGo results (titles, snippets, URLs) when the answer is not local

Pattern: summarize_corpus or list_documents → search_knowledge_base → read the exact section with get_document_outline + read_document_section before quoting.

Remember

• save_memory: a fact, preference or instruction to keep across sessions. Use when the user says "remember", or states a durable preference. One fact per call.
• search_memory: check before assuming; recalled memories are what was true when saved.
• Forgetting: delete_memory is not exposed by this server by default, because it deletes. Find the memory with search_memory, then delete it with DELETE /api/memory/<id> on the backend. An operator who wants the tool over MCP adds delete_memory to server.tools.allow in data/config/mcp.json.

Rules

• Quote the source (document name and section) for any claim that came from retrieval.
• If retrieval returns nothing, say so; do not fill the gap from general knowledge without labelling it.
• Indexing new folders happens in the Studio's Library page; the MCP server reads what is indexed.