Utility tools
A small set of tools that don't fit the design/run/manage categories elsewhere in this section — useful on their own, without needing an agent in the loop at all.
| Tool | What it does |
|---|---|
llm_chat | Sends a one-off prompt to a model through one of your project's configured LLM providers, without creating or running an agent |
analyze_uploaded_file | Extracts and returns the text content of a file previously uploaded to a chat session — reads from cache when available, otherwise extracts it fresh |
search_knowledge_base | Performs semantic search over your organization's knowledge base and returns matching passages |
Scoped to your organization
Like every other tool, these run against your project's own configured providers and your organization's own knowledge base — there's no cross-organization access.
When to reach for these
- Use
llm_chatfor a quick one-off completion when you don't need the full agent runtime — permission checks, tool dispatch, and conversation memory don't apply. - Use
analyze_uploaded_fileto pull text out of a file your MCP client already has a reference to, without re-uploading or re-parsing it yourself. - Use
search_knowledge_baseto check what your knowledge base actually returns for a query before wiring the same retrieval into a live agent.