Build a clinical agent

API reference Docs home

A Phenoml agent reasons over a conversation and calls Phenoml tools, your workflows, and registered MCP servers against a bound provider. You compose one from reusable prompts, tools, and provider bindings.

Watchdog

After the model acts, a watchdog evaluates whether the user's request is complete, needs more information, should retry with corrected input, or is blocked by a permanent limitation.

Sessions

Pass session_id to continue a conversation. Use streaming to render tokens and tool activity as they happen.

const agent = await client.agent.create({
  name: "Chart Q&A",
  prompts: ["prompt_cautious_clinician"],
  tools: ["mcp_scheduling"],
  provider: "550e8400-e29b-41d4-a716-446655440000",
});

const reply = await client.agent.chat({
  agent_id: agent.data.id,
  message: "What was this patient's most recent A1c, and is it controlled?",
  session_id: "session-abc123",
});

See the Agent reference.