Agent Crew API

The Agent Crew API provides endpoints for kicking off multi-agent "crews", monitoring their progress, and retrieving historical run data.

Endpoints

Run a Crew

POST /agents/crew/run

Dispatches a new agent crew task.

Request Body:

• prompt (string, required): The research prompt for the crew.
• session_id (string, optional): A unique identifier for the chat session.
• config_path (string, optional): Path to a specific agent configuration YAML.

Response (TaskAccepted):

• task_id (string): The Celery task ID for the background run.
• stream_url (string): The URL for the WebSocket event stream.

---

List Crew Runs

GET /agents/crews

Lists historical crew runs.

Query Parameters:

• limit (integer, default: 50): Number of runs to return.
• status (string, optional): Filter by run status (e.g., PENDING, STARTED, SUCCESS, FAILURE).

Response: A list of CrewRunSummary objects.

---

Get Crew Run Detail

GET /agents/crews/{task_id}

Retrieves detailed information and cached events for a specific crew run.

Response: A CrewRunDetail object containing the run summary, result, and event log.

---

Get Crew Events

GET /agents/crews/{task_id}/events

Returns an ordered stream of event frames for the task.

Query Parameters:

• limit (integer, default: 200): Number of recent events to return.

---

List Agent Tools

GET /agents/tools

Returns a registry of all available tools that agents can use, including their schemas and documentation.

Schemas

CrewRunSummary

Field	Type	Description
id	string	Internal database ID
task_id	string	Celery task ID
crew_name	string	Name of the crew configuration used
status	string	Current status of the run
prompt	string	The original prompt
started_at	datetime	When the run started
completed_at	datetime	When the run finished
cost_usd	float	Estimated token cost in USD

CrewRunDetail (extends CrewRunSummary)

Field	Type	Description
result	dict	The final output of the crew
events	list	Full ordered list of event frames