Skip to main content

Tasks

A task is a unit of work assigned to the AI agent. Each prompt you send creates a task that the agent processes inside a running workspace.

Task Lifecycle

Every task moves through a defined set of states:
StateDescription
createdTask is queued with your prompt
runningAgent is actively working and streaming output
waitingAgent asked a question and is waiting for your response
completedAgent finished the task successfully
failedAgent encountered an unrecoverable error
abortedYou cancelled the task before completion

Creating Tasks

Open your project and type a prompt in the chat input:
Add a dark mode toggle to the navbar using Tailwind’s dark: variant
The agent starts immediately and streams its work into the chat view.

Watching Task Output

If you created a task with task create or want to reconnect to an ongoing task:
aether task watch <task-id>
This streams the agent’s output in real-time, including text, tool invocations, and results.

Responding to Questions

When the agent needs clarification, the task enters waiting state. The agent presents one or more questions.
Questions appear inline in the chat. Type your response and press Enter.
If you’re using aether run, questions are handled interactively — they appear in your terminal and you can respond directly.

Aborting Tasks

To cancel a running task:
Click the Stop button in the chat view.
Aborting stops the agent immediately. Any files already written or commands already executed are not rolled back.

Listing Tasks

# List all tasks for the current project
aether task list

# Filter by status
aether task list --status running
Output:
ID          STATUS      PROMPT                          CREATED
task-001    completed   Add dark mode toggle            2025-03-15 14:30
task-002    running     Fix login page validation       2025-03-15 15:10
task-003    waiting     Set up PostgreSQL database      2025-03-15 15:25

Event Types

As a task runs, the agent produces a stream of typed events:
EventDescription
textAgent’s text output (reasoning, explanations)
thinkingInternal reasoning steps (model thinking)
tool_invocationA tool being called (file write, bash command, browser action)
resultThe result of a tool invocation
questionAgent is asking for user input
logSystem-level log messages
errorAn error occurred during execution
doneTask completed
finish_taskAgent signaled task completion
In the web UI, tool invocations are rendered with expandable details showing the tool name, input parameters, and output. In the CLI, they stream as formatted text.