Configure what the agent does automatically — daily routines, catch-ups, goal checks, knowledge contributions, and escalation behavior.
Autonomy settings are the agent's daily routine. Just like a good employee doesn't need to be told to check their email every morning, an autonomous agent can clock in on schedule, review unread messages, check on goals, and escalate issues to its manager — all without being asked. This tab lets you configure exactly which automatic behaviors are enabled and when they happen.
Autonomy settings are stored in engineAgent.config.autonomy. The save mechanism is context-aware:
POST /agents/:id/hot-update for live config updates. The agent reloads within 10 minutes.PATCH /agents/:id/config for static config updates. Takes effect on next start.The form tracks dirty state and only enables the Save button when changes exist. All times are interpreted in the agent's configured timezone (from the Workforce tab schedule).
engineAgent.config.autonomy and merged with defaults.The "Enable Autonomy System" toggle is the master switch for all automated behaviors. When disabled, no autonomous actions will be taken regardless of individual toggle settings. All other cards show reduced opacity when the master switch is off.
Automatically clocks the agent in and out based on the work schedule defined in the Workforce tab. No additional configuration needed here — it uses the schedule times directly.
At a set time each workday, the agent reviews unread emails, messages, and pending tasks, then sends a summary to its manager. The time is configured in the Manager & Catch-Up tab.
A deeper weekly review that includes summaries, reports, and planning. Configure which day of the week to send it using the day-of-week selector. Uses the same time as the daily catch-up.
The agent reviews Google Tasks at configured hours. Enter check times as comma-separated 24-hour values (e.g., "14, 17" for 2 PM and 5 PM). The last check of the day serves as an end-of-day review.
The agent reviews its learnings from the week and contributes to the organization's role-based knowledge base. Configure the day and time for this contribution.
When the agent can't answer a question, it follows a three-step process:
This prevents the agent from giving wrong answers when it doesn't know something.
When enabled, the agent self-monitors by evaluating guardrail rules on inbound emails and tool calls at runtime. This is a proactive safety measure — the agent checks itself before acting.
When the agent can't access a needed file (permissions error), it requests access from its manager instead of failing silently. This ensures the agent gets the resources it needs while maintaining access controls.
This is normal for running agents. Hot updates are queued and applied at the next configuration check cycle (up to 10 minutes). If you need immediate effect, restart the agent.
Check: (1) The master switch is enabled, (2) Daily catch-up is toggled on, (3) The agent is clocked in at the catch-up time, (4) The agent has a manager configured. Also check the Activity tab for any errors.
Verify the goal check hours are valid 24-hour values (0-23) separated by commas. The agent must be online and clocked in at the check time. Google Tasks integration must be configured.
You have modified settings that haven't been saved. Click "Save" to persist them or reload the page to discard. The dirty state tracker compares current form values against the loaded config.