Skip to main content
The built-in agent answers questions about the selected person’s health record. Use it in the web client at /ask or through the chat API. To use Claude Code, Codex, Cursor or another external agent, connect through MCP.

Available tools

Citations link an answer to a recorded reading, a document passage or a medical reference. Open a citation to check the source behind the answer. Reasoning and tool activity appear in the work panel; the final answer appears separately. Long conversations are summarized automatically as they approach the context limit. The panel shows when summarization is in progress. Stop generating ends reception in the browser. Generation may continue on the server, so a completed answer can appear when you refresh the conversation.

Configuration

Choose a model in Settings › Model or follow Configuration. The main agent settings are: Knowledge tools appear only when a local or online source is available. Calculation needs no setup with a cloud model; a local model calculates in Python when the app can reach a Docker daemon, and in JavaScript otherwise. mirobody doctor --probe checks model capabilities and reports whether the endpoint showed its reasoning; a model that thinks adaptively may skip it on the probe’s short question, which is a note, not a failure.

Custom agents

A deployment can replace the built-in agent through AGENT_DIRS or an installed package’s mirobody.agents entry point. To reuse the harness in another application, install mirobody[agent]. The integration contract and example plugin are in .