Agents & tools

mistral.rs can run the entire tool-calling loop server-side, execute Python, shell commands, and web searches locally, use OpenAI-compatible Skills, accept OpenAI-compatible file inputs, and stream model text, tool progress, files, and media from one request.

The agent system has three layers:

• Tool protocol - how a model requests external work.
• Tool execution - built-in code execution, shell, web search, OpenAI-compatible Skills, MCP (Model Context Protocol) tools, callbacks, or external dispatch.
• App runtime - streaming output, tool progress, files, generated media, and sessions as one local surface.

Where to start:

• Tool calling for the protocol.
• Build an agent for the end-to-end walkthrough.
• OpenAI-compatible Skills for uploaded Skills and local SDK mounts.
• OpenAI-compatible file inputs for uploaded, inline, and URL-backed request files.
• Shell execution for command execution, working directories, and sandboxing.