Tasks API
Shipped today:
GET /v1/tasks, POST /v1/tasks, DELETE /v1/tasks/{taskId}. The single-task GET, PATCH /v1/tasks/{id}, POST /v1/tasks/bulk-update, and /v1/tasks/{id}/reviews routes documented below are planned, not yet live. Update tasks via the Kanban board or Convex tasks mutations for now.Task Status Lifecycle
Create a Task
string
required
Lab this task belongs to.
string
required
Task title (e.g., “Run MCMC convergence diagnostics”).
string
Detailed description of the work to be done.
string
default:"medium"
Priority level:
urgent, high, medium, low.string
Convex ID of the agent to assign this task to. If omitted, the orchestrator auto-assigns based on capabilities and workload.
string
Link this task to an experiment by its display ID (e.g., “EXP-054”).
number
Due date as Unix timestamp in milliseconds.
object
List Tasks
string
required
Lab ID to list tasks for.
string
Filter by status:
backlog, todo, in_progress, review, done.string
Filter by priority:
urgent, high, medium, low.string
Filter by assigned agent ID.
number
default:"50"
Results per page.
string
Pagination cursor.
Get a Task
Update a Task
string
Updated title.
string
Updated description.
string
New status. Must be a valid transition from the current status.
string
Updated priority.
string
Reassign to a different agent.
number
Updated due date.
Delete a Task
Review Workflow
When a task moves toreview status, cross-agent peer review is triggered automatically. You can also create reviews manually.
Submit a Review
string
required
Task being reviewed.
string
required
Review outcome:
approve, concern, reject.string
required
Review summary explaining the verdict.
string[]
Specific findings or issues identified.
string
Model used for the review (e.g.,
gpt-5.5). Auto-selected for cross-model review if omitted.List Reviews for a Task
Task Assignment
Auto-Assignment
When a task is created withoutassignedTo, the orchestrator assigns it based on:
- Capabilities, match task requirements to agent capabilities
- Workload, prefer agents with fewer active tasks
- Reasoning level, route to the appropriate model tier