Skip to main content

Tasks

A task is a natural language instruction that Miteos decomposes and executes via one or more agents.

Creating a Task

Submit a task via the War Room prompt bar or the API:
POST /tasks
{
  "prompt": "Research the top 5 SaaS competitors to Notion and create a feature comparison spreadsheet",
  "workspace_id": "optional-workspace-id"
}

Task Decomposition

Miteos automatically breaks complex tasks into subtasks:
"Research competitors and build a pricing dashboard"
├── Subtask 1: Research competitor pricing pages (researcher)
├── Subtask 2: Extract and structure pricing data (analyst)
└── Subtask 3: Build interactive dashboard component (builder)
Each subtask is assigned to the most appropriate agent type based on the required tools and capabilities.

Task States

StateDescription
pendingTask created, waiting to start
runningAgents actively working
completedAll subtasks finished successfully
failedOne or more subtasks failed unrecoverably
cancelledUser cancelled the task

Monitoring

Track task progress in real-time:
GET /tasks/{task_id}
Returns:
{
  "id": "task-abc123",
  "prompt": "Research competitor pricing...",
  "status": "running",
  "total_tokens_used": 4521,
  "total_cost_cents": 2,
  "created_at": "2026-06-20T10:00:00Z",
  "subtasks": [
    {"id": "st-1", "agent_type": "researcher", "status": "complete"},
    {"id": "st-2", "agent_type": "analyst", "status": "running"},
    {"id": "st-3", "agent_type": "builder", "status": "pending"}
  ]
}

Cancellation

Cancel a running task (kills all active agents):
POST /tasks/{task_id}/cancel

Deliverables

When a task completes, its outputs are available as deliverables:
GET /tasks/{task_id}/deliverables
Each deliverable can be:
  • Downloaded as a file
  • Deployed to a staging URL
  • Shared via a public link

Task History

List all your tasks:
GET /tasks?limit=20&offset=0

Completed Tasks (for Skill Compiler)

Successfully completed tasks can be used to compile reusable skills:
GET /tasks/completed
The Skill Compiler analyzes the action patterns from successful tasks and extracts them into named, reusable skills that execute faster on repeat.