Approvals & autonomous mode
Silico is built so an agent working on its own can't do anything irreversible without your say-so — and so that, when you're ready to trust one, you can safely take the training wheels off. This page covers both sides: the approval gates that keep unattended work safe by default, and autonomous mode, the opt-in that lets a trusted automation skip the prompts.
Approval gates (the safe default)
When a background job wants to do something irreversible or outward-facing — run a shell command, execute code, click or type in a real browser, or write through a connected tool — it doesn't just do it. It pauses and asks you first.
You'll see the job go to Paused · approval needed in Activity. Expanding it shows exactly what it wants to do and two buttons:
- Approve — run this one action.
- Reject — skip it and continue.
Approving clears that single action only — it's not a standing permission. The next risky action pauses again. If you're away, Silico emails you a "while you were away" digest summarizing what happened, what ran, and what's still waiting — and you can approve right from the email.
Harmless, reversible actions (like reading a file or writing to the job's own scratch space) don't interrupt you; only the genuinely risky ones do.
Autonomous mode
For an automation you've come to trust, you can turn on Autonomous mode so it stops pausing for approval and simply runs its actions. It's the opt-out of the prompts above — and only that. Every other safety brake stays on.
Turning it on
Open Automations → (pick one) → manage. At the top you'll find a switch:
Autonomous mode — Run without approval prompts
The first time you enable it, a confirmation appears:
"Autonomous mode removes approval prompts — this job can run irreversible actions (shell, browser) on its own. Budget, stall, and safety limits still apply."
Choose Enable autonomous mode to confirm, or Cancel. Nothing turns on until you explicitly confirm — and after that first time, toggling it later is just a normal switch.
What still protects you
Autonomous mode removes approval prompts, and nothing else. All of these stay fully in force:
- Spend limits still apply. The job's budget ceiling still warns and still pauses at the cap. Autonomy never means unlimited spend.
- Stall and safety limits still apply. A stuck or looping job is still caught and paused; failing jobs still stop.
- Unfamiliar tools still pause. Auto-approval only covers Silico's known, built-in actions. Anything unusual — and any write through a connected tool — still stops and asks. It fails safe, not open.
- Everything is still recorded. Actions the automation ran on its own are logged and clearly labeled auto-approved, and appear in your digest under "ran autonomously per your setting." You always have a full, honest record of what it did.
Changing your mind
- Tighten a running job back to asking. You can switch a live autonomous job back to requiring approvals at any time. It's a one-way safety move — you can always de-escalate a running job to safe, but you can't flip a running job into autonomy mid-flight. Actions already taken aren't undone; the next risky one simply pauses again.
- Turn it off entirely. Flip the automation's toggle back off and it returns to asking for approval on everything.
How to think about it
Autonomous mode is best for routine, well-scoped automations you've watched succeed a few times — the "Monday competitor scan," the "nightly report," the "triage new signups." Keep approval gates on for anything new, high-stakes, or acting in your connected accounts, until you've seen it behave.
Autonomous mode is off by default everywhere. It's always something you choose, per automation, with a recorded confirmation — never a default and never a surprise.