| max_completion_tokens | number | No | Maximum output tokens. |
| max_iterations | number | No | Max reasoning/tool-use loops. |
| memory | object | No | How much history the agent sends with each request: {"kind": "window", "context_length": N} keeps the last N messages, {"kind": "off"} keeps none. Which memory is read is set per run or per step, never on the agent. |
| output_schema | object | No | Structured-output JSON schema. |
| output_type | string | No | |
| provider | object | Yes | AI provider + model + credentials resource for the agent. |
| streaming | boolean | No | |
| system_prompt | string | No | System prompt for the agent. |
| temperature | number | No | Sampling temperature (0.0-2.0). |
| tools | array | No | Reusable tool definitions available to the agent. |