The Turn object

A run has one turn per model call. A turn that decides to use a tool produces tool calls; the results go back to the model and the next turn begins. Turns are not addressable on their own — they appear in the turns array of GET /v1/loom/runs/:id/trace.

{
  "seq": 2,
  "model_class": "standard",
  "input_tokens": 1840,
  "output_tokens": 96,
  "started_at": "2026-08-30T12:00:02.140Z",
  "duration_ms": 1420,
  "cost_credits": 0.06,
  "request": { },
  "wire_request": { },
  "response": { },
  "tool_calls": []
}

Attributes

integer

required

Position within the run, starting at 1.

string

fast, standard, or frontier — which class actually served this call.

integer

Tokens sent. Usually the reason a run cost more than expected.

integer

Tokens generated.

string

ISO 8601 with milliseconds.

integer

Wall time for this model call.

number

Credits charged for this turn alone. The run totals them.

object

What we assembled for the model — instructions, context, tool definitions.

object

The provider-shaped payload actually sent. Useful when request looks right but behaviour does not.

object

What the model returned.

array

Tool calls this turn decided to make.

Core

Conversations

Configuration

Delivery

MCP

Workbench

Attributes

​Attributes

Attributes