Tasks Overview

Tasks are the fundamental unit of work in Agent HQ. Every piece of work — whether it’s writing code, fixing a bug, or running tests — is represented as a task.

What is a task?

A task is a discrete piece of work that Pilot executes autonomously. Each task has:

• Title — a short description of what needs to be done
• Body — detailed requirements or context
• Status — where the task is in its lifecycle
• Type — Code, Writing, or Thinking
• Model — which AI model to use
• Execution logs — real-time output from the running agent
• Cost — token usage and dollar cost
• Files — code and assets generated

Task types

Code

The most common type. Pilot writes, modifies, or refactors code. Code tasks have access to the sandbox environment with terminal, filesystem, and browser capabilities.

Writing

For content generation — documentation, README files, blog posts, or any text-based output.

Thinking

For analysis, planning, and reasoning tasks that don’t produce code. Useful for architectural decisions, code reviews, or breaking down complex problems.

How tasks are created

Through conversations

The most natural way. Chat with Pilot about what you need, and it creates tasks automatically:

You: Fix the 404 error on the /settings page Pilot: I’ll create a task to investigate and fix the routing issue.

From the task board

Click the + button on any column in the task board to create a task manually. Fill in the title, body, type, and model.

Through the API

Create tasks programmatically via the Tasks API.

Task execution

When a task runs, it executes in an isolated sandbox:

1. The project’s GitHub repo is cloned
2. A new branch is created for the task
3. Pilot reads the codebase and executes the requested changes
4. Code is written, tests are run, and results are captured
5. Changes are committed and pushed
6. A pull request is created if applicable

See Execution & Monitoring for details on watching tasks run.