Connect any REST API to an AI agent

What you can build

How to connect REST APIs

1. 01Create a tool of type HTTP API. Write the description for the model: what the endpoint does, what it needs, what it returns and when to use it.
2. 02Set the method and URL; use {{placeholders}} in the URL, query parameters, headers and body template. Each placeholder becomes a field the model must supply, described in the input schema.
3. 03Pick the auth type: API key, bearer token, basic, OAuth2, a custom header, a stored connection, or end user token to act as the signed-in user.
4. 04Restrict allowed domains, set the expected status codes, a timeout and a retry policy, and declare the safety level. Mark writes as requiring approval if a human must confirm them.
5. 05Test the tool from the toolbox, attach it to an agent and check the call in the run trace.

Example tool definition

The shape a tool takes in the AgentRow builder API. Credentials are stored on the tool's auth settings, not in its configuration, so they never reach a prompt. Placeholders in double braces are filled by the model from the input schema.

{
  "name": "get_order_status",
  "type": "http_api",
  "description": "Look up an order by id and return its status, items and delivery estimate. Use when the user asks about an order.",
  "auth_type": "api_key",
  "safety_level": "safe_read",
  "is_readonly": true,
  "timeout_ms": 8000,
  "config": {
    "method": "GET",
    "url": "https://api.example.com/v1/orders/{{order_id}}",
    "query_params": { "expand": "items" },
    "allowed_domains": ["api.example.com"],
    "expected_status_codes": [200, 404]
  },
  "input_schema": {
    "type": "object",
    "properties": {
      "order_id": { "type": "string", "description": "The order id, for example ORD-10422" }
    },
    "required": ["order_id"]
  }
}

What the run trace shows

Every call appears in the run trace with the rendered request, the response status and body, timing, retries and errors, plus what the model was shown of the result. If the model never called the tool, the toolbox step shows whether it was bound for that model call and why.