Initiates a new agent task execution.

This endpoint starts an agent that will perform the specified task. The agent can optionally navigate to a provided URL before attempting to solve the task.

This agent is non-blocking and will return immediately. You should periodically check the status of the agent until it is closed.

Important points:

• Remember that you can attach a vault_id to the agent to allow it to access credentials stored in your vault.
• Sometimes LLM providers have outages or cluster issues. Try to switch to a different reasoning model if that occurs.
• Not all models support vision. Check litellm documentation for more details.
• You can debug your agent execution by checking the replay endpoint.