Complete Your First Task

This exercise does not require a project and will not modify files. Use it to confirm that the model connection, message delivery, and task stopping all work as expected.

1. Confirm that a model is available

If you have not configured a model:

1. Select the profile area at the bottom of the left sidebar, then select “Settings.”
2. Open “Configuration” and enter the API URL and API token.
3. Confirm that at least one model is enabled. Its identifier must be a value that your service actually accepts.
4. Select “Back to app.”

Do not paste a token into a chat message. Connection settings and chat messages cross different data boundaries.

2. Start a new conversation

1. Select “New conversation” at the top of the left sidebar.
2. For the project, select “No project.”
3. Select the model you just configured.
4. Keep the permission mode set to Default.

Default permissions do not interfere with ordinary questions and answers, while helping prevent a task from unintentionally accessing files outside the workspace or automatically performing writes.

3. Send a verifiable request

Enter the following in the composer:

In three points, explain the difference between a standard chat model and an agent. For each point, state the conclusion first, then give an everyday work example. Do not call external tools.

Press Enter to send. You should see the agent begin processing and eventually return a structured response.

If no response arrives:

• Check the model service URL, token, and model identifier.
• Confirm that your network can reach the service.
• Determine whether the error concerns authentication, quota, service availability, or an incompatible response.
• Correct the configuration in Settings, then create a new task or retry.

4. Learn how the composer behaves during a run

While the agent is running:

• New text sent from the composer first enters the pending-message queue.
• You can reorder, edit, or remove queued messages. “Steer” passes an appropriate message into the current run.
• When the composer is empty, select the stop button to request that the current run terminate.

Stopping is a request, not a transactional rollback. If a tool has already written a file, run a command, or sent a request to a third party, inspect the actual result.

5. Manage the result

After the task finishes, you can:

• Copy or bookmark a message.
• Edit the last user message and rerun that turn.
• Select “Continue in a new task” from a completed assistant response to create a new conversation with a history snapshot.
• Pin, rename, or archive the conversation in the left sidebar.

Next step

When you are ready to let an agent read real files, continue to Use Your First Project.