# Start an agent run

Starts the model working on the conversation's next turn and answers 202 with the run's id at once; the work happens after the response. A client-chosen `runId` makes the call safe to retry: calling it again with the same id — whether the first call already created the run or is still being admitted — answers the same id rather than starting a second run. The run's progress and output follow on the journal and `runs.get`; a tool call that needs consent opens an approval and pauses the run until `approvals.approve` or `approvals.deny` decides it.

`POST /v1/users/{userId}/agent-runs`

## Authentication

Both headers are required.

- Header: `X-Api-Key: YOUR_API_KEY`
- Header: `Authorization: Bearer YOUR_USER_TOKEN`

## Path parameters



```json
[
  {
    "schema": {
      "type": "string",
      "format": "uuid",
      "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$"
    },
    "in": "path",
    "name": "userId",
    "required": true,
    "description": "The user in the path; must equal the token's own account."
  }
]
```


## Request body



```json
{
  "required": true,
  "content": {
    "application/json": {
      "schema": {
        "type": "object",
        "properties": {
          "conversationId": {
            "type": "string",
            "format": "uuid",
            "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$",
            "description": "The conversation's id."
          },
          "text": {
            "type": "string",
            "minLength": 1,
            "maxLength": 32768,
            "description": "The run's seed instruction, up to 32768 characters."
          },
          "profile": {
            "type": "object",
            "properties": {
              "tools": {
                "description": "The only tools this run may reach. Absent means every registered tool.",
                "maxItems": 64,
                "type": "array",
                "items": {
                  "type": "string",
                  "pattern": "^[a-z][a-z0-9_]{0,63}$",
                  "description": "A tool's name: lowercase, starting with a letter, letters, digits and underscores only."
                }
              },
              "outputContract": {
                "description": "The strict terminal output contract validated and frozen during admission.",
                "type": "object",
                "properties": {
                  "name": {
                    "type": "string",
                    "pattern": "^[a-z][a-z0-9_]{0,63}$",
                    "description": "A tool's name: lowercase, starting with a letter, letters, digits and underscores only."
                  },
                  "description": {
                    "description": "What the terminal result means, shown beside the run's output.",
                    "type": "string",
                    "maxLength": 4096
                  },
                  "schema": {
                    "description": "A bounded, strict JSON Schema object (`type: \"object\"`, `additionalProperties: false`) the run's terminal output must satisfy."
                  }
                },
                "required": [
                  "name",
                  "schema"
                ],
                "additionalProperties": false
              }
            },
            "additionalProperties": false,
            "description": "Bounds on this run beyond its seed text: which tools it may reach, and its terminal output contract."
          },
          "runId": {
            "description": "A client-chosen id that makes the start safe to retry: calling this again with the same id — whether the first call already created the run or is still being admitted — answers the same run rather than starting a second one.",
            "type": "string",
            "format": "uuid",
            "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$"
          }
        },
        "required": [
          "conversationId",
          "text"
        ],
        "additionalProperties": false
      }
    }
  }
}
```


## Request example

Illustrative request. Replace the host, credentials and resource IDs with your own. If a request file is shown, create it from the schema above. Review the requested action before sending it.

```sh
curl --request POST 'https://api.example.test/v1/users/YOUR_USER_ID/agent-runs' \
  --header 'X-Api-Key: YOUR_API_KEY' \
  --header 'Authorization: Bearer YOUR_USER_TOKEN' \
  --header 'Content-Type: application/json' \
  --data-binary '@request.json'
```

## Responses

### 202

The run's id; the same id again when this exact `runId` was already accepted.



```json
{
  "description": "The run's id; the same id again when this exact `runId` was already accepted.",
  "content": {
    "application/json": {
      "schema": {
        "type": "object",
        "properties": {
          "runId": {
            "type": "string",
            "format": "uuid",
            "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$",
            "description": "The started run's id; the same value sent, when one was sent."
          }
        },
        "required": [
          "runId"
        ]
      }
    }
  }
}
```


### 400

The request could not be read as this operation expects.

- `invalid_input` — the body or query failed validation; `issues` names each field
- `invalid_cursor` — the `cursor` is not one this list minted



```json
{
  "description": "The request could not be read as this operation expects.\n\n- `invalid_input` — the body or query failed validation; `issues` names each field\n- `invalid_cursor` — the `cursor` is not one this list minted",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 401

A credential is missing, invalid or expired.

- `partner_key_required` — no `X-Api-Key` header was sent
- `partner_key_invalid` — the `X-Api-Key` is unknown or revoked
- `credential_expired` — the user token has expired; obtain a fresh one



```json
{
  "description": "A credential is missing, invalid or expired.\n\n- `partner_key_required` — no `X-Api-Key` header was sent\n- `partner_key_invalid` — the `X-Api-Key` is unknown or revoked\n- `credential_expired` — the user token has expired; obtain a fresh one",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 403

The credentials are valid but may not do this.

- `subject_mismatch` — the `{userId}` in the path is not the token's user
- `account_disabled` — an operator disabled the account
- `origin_rejected` — a browser `Origin` other than the configured web origin



```json
{
  "description": "The credentials are valid but may not do this.\n\n- `subject_mismatch` — the `{userId}` in the path is not the token's user\n- `account_disabled` — an operator disabled the account\n- `origin_rejected` — a browser `Origin` other than the configured web origin",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 404

The resource is missing, belongs to someone else, or its id is malformed: all three answer alike.



```json
{
  "description": "The resource is missing, belongs to someone else, or its id is malformed: all three answer alike.",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 409

The operation lost to the current state.

- `run_active` — a run already holds this conversation
- `budget_exhausted` — the user's spend headroom is gone, or an operator froze it
- `execution_capacity` — no execution capacity is free right now; `retryable` says whether to try again
- `execution_unavailable` — the execution engine could not take the work
- `profile_unknown_tool` — the run profile names a tool this deployment does not have
- `deployment_paused` — an operator paused a deployment control; nothing was admitted or fired
- `stop_pending` — a stop is already in progress and its cleanup is not yet confirmed



```json
{
  "description": "The operation lost to the current state.\n\n- `run_active` — a run already holds this conversation\n- `budget_exhausted` — the user's spend headroom is gone, or an operator froze it\n- `execution_capacity` — no execution capacity is free right now; `retryable` says whether to try again\n- `execution_unavailable` — the execution engine could not take the work\n- `profile_unknown_tool` — the run profile names a tool this deployment does not have\n- `deployment_paused` — an operator paused a deployment control; nothing was admitted or fired\n- `stop_pending` — a stop is already in progress and its cleanup is not yet confirmed",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 412

The identity is verified but has no account yet.

- `account_not_provisioned` — the identity is verified but has no account yet; call `users.ensure` first



```json
{
  "description": "The identity is verified but has no account yet.\n\n- `account_not_provisioned` — the identity is verified but has no account yet; call `users.ensure` first",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 413

The body exceeds the size this deployment accepts.



```json
{
  "description": "The body exceeds the size this deployment accepts.",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 429

A limiter refused the request; honour `Retry-After`.

- `rate_limited` — the per-key or per-user limit is spent; honour `Retry-After`



```json
{
  "description": "A limiter refused the request; honour `Retry-After`.\n\n- `rate_limited` — the per-key or per-user limit is spent; honour `Retry-After`",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 500

A fault on our side; quote `requestId` when reporting it.

- `internal` — a fault on our side; quote `requestId` when reporting it



```json
{
  "description": "A fault on our side; quote `requestId` when reporting it.\n\n- `internal` — a fault on our side; quote `requestId` when reporting it",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


### 503

The replica is draining or a dependency did not answer; `retryable` says whether to try again.

- `internal` — a fault on our side; quote `requestId` when reporting it



```json
{
  "description": "The replica is draining or a dependency did not answer; `retryable` says whether to try again.\n\n- `internal` — a fault on our side; quote `requestId` when reporting it",
  "content": {
    "application/problem+json": {
      "schema": {
        "$ref": "#/components/schemas/ProblemDetails"
      }
    }
  }
}
```


## Complete OpenAPI definition

The exact operation and all referenced components, including recursive schemas.



```json
{
  "operation": {
    "operationId": "runs.start",
    "summary": "Start an agent run",
    "tags": [
      "agent-runs"
    ],
    "description": "Starts the model working on the conversation's next turn and answers 202 with the run's id at once; the work happens after the response. A client-chosen `runId` makes the call safe to retry: calling it again with the same id — whether the first call already created the run or is still being admitted — answers the same id rather than starting a second run. The run's progress and output follow on the journal and `runs.get`; a tool call that needs consent opens an approval and pauses the run until `approvals.approve` or `approvals.deny` decides it.",
    "requestBody": {
      "required": true,
      "content": {
        "application/json": {
          "schema": {
            "type": "object",
            "properties": {
              "conversationId": {
                "type": "string",
                "format": "uuid",
                "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$",
                "description": "The conversation's id."
              },
              "text": {
                "type": "string",
                "minLength": 1,
                "maxLength": 32768,
                "description": "The run's seed instruction, up to 32768 characters."
              },
              "profile": {
                "type": "object",
                "properties": {
                  "tools": {
                    "description": "The only tools this run may reach. Absent means every registered tool.",
                    "maxItems": 64,
                    "type": "array",
                    "items": {
                      "type": "string",
                      "pattern": "^[a-z][a-z0-9_]{0,63}$",
                      "description": "A tool's name: lowercase, starting with a letter, letters, digits and underscores only."
                    }
                  },
                  "outputContract": {
                    "description": "The strict terminal output contract validated and frozen during admission.",
                    "type": "object",
                    "properties": {
                      "name": {
                        "type": "string",
                        "pattern": "^[a-z][a-z0-9_]{0,63}$",
                        "description": "A tool's name: lowercase, starting with a letter, letters, digits and underscores only."
                      },
                      "description": {
                        "description": "What the terminal result means, shown beside the run's output.",
                        "type": "string",
                        "maxLength": 4096
                      },
                      "schema": {
                        "description": "A bounded, strict JSON Schema object (`type: \"object\"`, `additionalProperties: false`) the run's terminal output must satisfy."
                      }
                    },
                    "required": [
                      "name",
                      "schema"
                    ],
                    "additionalProperties": false
                  }
                },
                "additionalProperties": false,
                "description": "Bounds on this run beyond its seed text: which tools it may reach, and its terminal output contract."
              },
              "runId": {
                "description": "A client-chosen id that makes the start safe to retry: calling this again with the same id — whether the first call already created the run or is still being admitted — answers the same run rather than starting a second one.",
                "type": "string",
                "format": "uuid",
                "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$"
              }
            },
            "required": [
              "conversationId",
              "text"
            ],
            "additionalProperties": false
          }
        }
      }
    },
    "parameters": [
      {
        "schema": {
          "type": "string",
          "format": "uuid",
          "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$"
        },
        "in": "path",
        "name": "userId",
        "required": true,
        "description": "The user in the path; must equal the token's own account."
      }
    ],
    "security": [
      {
        "apiKey": [],
        "userToken": []
      }
    ],
    "responses": {
      "202": {
        "description": "The run's id; the same id again when this exact `runId` was already accepted.",
        "content": {
          "application/json": {
            "schema": {
              "type": "object",
              "properties": {
                "runId": {
                  "type": "string",
                  "format": "uuid",
                  "pattern": "^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$",
                  "description": "The started run's id; the same value sent, when one was sent."
                }
              },
              "required": [
                "runId"
              ]
            }
          }
        }
      },
      "400": {
        "$ref": "#/components/responses/InvalidInput"
      },
      "401": {
        "$ref": "#/components/responses/Unauthorized"
      },
      "403": {
        "$ref": "#/components/responses/Forbidden"
      },
      "404": {
        "$ref": "#/components/responses/NotFound"
      },
      "409": {
        "description": "The operation lost to the current state.\n\n- `run_active` — a run already holds this conversation\n- `budget_exhausted` — the user's spend headroom is gone, or an operator froze it\n- `execution_capacity` — no execution capacity is free right now; `retryable` says whether to try again\n- `execution_unavailable` — the execution engine could not take the work\n- `profile_unknown_tool` — the run profile names a tool this deployment does not have\n- `deployment_paused` — an operator paused a deployment control; nothing was admitted or fired\n- `stop_pending` — a stop is already in progress and its cleanup is not yet confirmed",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "412": {
        "$ref": "#/components/responses/AccountNotProvisioned"
      },
      "413": {
        "$ref": "#/components/responses/PayloadTooLarge"
      },
      "429": {
        "$ref": "#/components/responses/TooManyRequests"
      },
      "500": {
        "$ref": "#/components/responses/Internal"
      },
      "503": {
        "$ref": "#/components/responses/Unavailable"
      }
    }
  },
  "components": {
    "responses": {
      "InvalidInput": {
        "description": "The request could not be read as this operation expects.\n\n- `invalid_input` — the body or query failed validation; `issues` names each field\n- `invalid_cursor` — the `cursor` is not one this list minted",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "Unauthorized": {
        "description": "A credential is missing, invalid or expired.\n\n- `partner_key_required` — no `X-Api-Key` header was sent\n- `partner_key_invalid` — the `X-Api-Key` is unknown or revoked\n- `credential_expired` — the user token has expired; obtain a fresh one",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "Forbidden": {
        "description": "The credentials are valid but may not do this.\n\n- `subject_mismatch` — the `{userId}` in the path is not the token's user\n- `account_disabled` — an operator disabled the account\n- `origin_rejected` — a browser `Origin` other than the configured web origin",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "NotFound": {
        "description": "The resource is missing, belongs to someone else, or its id is malformed: all three answer alike.",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "AccountNotProvisioned": {
        "description": "The identity is verified but has no account yet.\n\n- `account_not_provisioned` — the identity is verified but has no account yet; call `users.ensure` first",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "PayloadTooLarge": {
        "description": "The body exceeds the size this deployment accepts.",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "TooManyRequests": {
        "description": "A limiter refused the request; honour `Retry-After`.\n\n- `rate_limited` — the per-key or per-user limit is spent; honour `Retry-After`",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "Internal": {
        "description": "A fault on our side; quote `requestId` when reporting it.\n\n- `internal` — a fault on our side; quote `requestId` when reporting it",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      },
      "Unavailable": {
        "description": "The replica is draining or a dependency did not answer; `retryable` says whether to try again.\n\n- `internal` — a fault on our side; quote `requestId` when reporting it",
        "content": {
          "application/problem+json": {
            "schema": {
              "$ref": "#/components/schemas/ProblemDetails"
            }
          }
        }
      }
    },
    "schemas": {
      "ProblemDetails": {
        "type": "object",
        "properties": {
          "type": {
            "type": "string",
            "description": "The kind of problem as a URN, `urn:fin:error:<kind>`; stable, compare against it"
          },
          "title": {
            "type": "string",
            "description": "The kind's human title, for logs; never parse it"
          },
          "status": {
            "type": "integer",
            "minimum": 400,
            "maximum": 599,
            "description": "The HTTP status, repeated in the body"
          },
          "reason": {
            "description": "The machine-readable why. One of:\n\n- `run_active` (409) — a run already holds this conversation\n- `budget_exhausted` (409) — the user's spend headroom is gone, or an operator froze it\n- `approval_not_pending` (409) — the approval was already decided or has expired, or its id does not exist or belongs to a different conversation\n- `execution_capacity` (409) — no execution capacity is free right now; `retryable` says whether to try again\n- `execution_unavailable` (409) — the execution engine could not take the work\n- `stop_pending` (409) — a stop is already in progress and its cleanup is not yet confirmed\n- `automation_changed` (409) — the `revision` sent is stale; reload the automation\n- `automation_held` (409) — an operator holds the automation; it fires again when released\n- `automation_invalid` (409) — the automation's definition cannot run as written\n- `automation_completed` (409) — the automation has finished for good and cannot fire again\n- `automation_limit` (409) — the user already has as many automations as the deployment allows\n- `profile_unknown_tool` (409) — the run profile names a tool this deployment does not have\n- `deployment_paused` (409) — an operator paused a deployment control; nothing was admitted or fired\n- `run_not_active` (409) — the run named in the path is not the conversation's live run\n- `withdrawal_changed` (409) — the withdrawal cannot be prepared or confirmed as asked: the balance no longer covers it, its terms changed or expired, or it is already in progress\n- `credential_expired` (401) — the user token has expired; obtain a fresh one\n- `account_disabled` (403) — an operator disabled the account\n- `account_not_provisioned` (412) — the identity is verified but has no account yet; call `users.ensure` first\n- `provider_unavailable` (503) — an external provider the call depends on did not answer\n- `engine_unavailable` (503) — the execution engine did not answer\n- `invalid_input` (400) — the body or query failed validation; `issues` names each field\n- `internal` (500) — a fault on our side; quote `requestId` when reporting it\n- `partner_key_required` (401) — no `X-Api-Key` header was sent\n- `partner_key_invalid` (401) — the `X-Api-Key` is unknown or revoked\n- `subject_mismatch` (403) — the `{userId}` in the path is not the token's user\n- `origin_rejected` (403) — a browser `Origin` other than the configured web origin\n- `permission_required` (403) — the operator credential lacks the scope this call needs\n- `rate_limited` (429) — the per-key or per-user limit is spent; honour `Retry-After`\n- `stream_capacity` (429) — no stream socket is free on this replica or for this user; honour `Retry-After`\n- `invalid_cursor` (400) — the `cursor` is not one this list minted",
            "type": "string",
            "enum": [
              "run_active",
              "budget_exhausted",
              "approval_not_pending",
              "execution_capacity",
              "execution_unavailable",
              "stop_pending",
              "automation_changed",
              "automation_held",
              "automation_invalid",
              "automation_completed",
              "automation_limit",
              "profile_unknown_tool",
              "deployment_paused",
              "run_not_active",
              "withdrawal_changed",
              "credential_expired",
              "account_disabled",
              "account_not_provisioned",
              "provider_unavailable",
              "engine_unavailable",
              "invalid_input",
              "internal",
              "partner_key_required",
              "partner_key_invalid",
              "subject_mismatch",
              "origin_rejected",
              "permission_required",
              "rate_limited",
              "stream_capacity",
              "invalid_cursor"
            ]
          },
          "requestId": {
            "type": "string",
            "description": "The id Fin used for this request; quote it when reporting a problem"
          },
          "retryable": {
            "type": "boolean",
            "description": "Whether repeating the same request later can succeed without changing it"
          },
          "detail": {
            "description": "Only on `invalid_input`: which part of the request failed validation",
            "type": "string"
          },
          "issues": {
            "description": "Only on `invalid_input`: one entry per failing field",
            "type": "array",
            "items": {
              "type": "object",
              "properties": {
                "path": {
                  "type": "string",
                  "description": "The JSON pointer of the failing field; empty for the root object"
                },
                "message": {
                  "type": "string",
                  "description": "Why the field failed"
                }
              },
              "required": [
                "path",
                "message"
              ]
            }
          }
        },
        "required": [
          "type",
          "title",
          "status",
          "requestId",
          "retryable"
        ],
        "description": "RFC 9457 problem details: what every error response carries. Never a provider's message, a query or a stack."
      }
    },
    "securitySchemes": {
      "apiKey": {
        "type": "apiKey",
        "in": "header",
        "name": "X-Api-Key"
      },
      "userToken": {
        "type": "http",
        "scheme": "bearer",
        "bearerFormat": "JWT"
      }
    }
  }
}
```

