Automations & background jobs
Not everything should happen while you watch. Silico can run work in the background and on a schedule — long tasks that keep going after you close the tab, and recurring tasks that run themselves.
Background jobs
When you send work that takes a while — a big research task, a rendering job, a scheduled run — Silico turns it into a background job. It keeps running on the server even if you close the tab or lose your connection, and picks up where it left off if anything restarts. You're never charged twice for a step it already finished.
The Activity panel
Open Activity (the inbox icon in the left sidebar, with an unread badge) to see all your background jobs — running, paused, and finished — newest first. It refreshes live while work is in progress.
Each job shows a status:
| Status | What it means |
|---|---|
| Running | Actively working (step N · running) |
| Done | Finished — expand to see the result and files |
| Paused · approval needed | Waiting for you to approve a risky action |
| Paused · cap reached | Hit its spend limit — raise the cap or wait |
| Failed | Errored out, with a reason |
| Queued | Waiting for a free runner |
Expand any finished job to see the deliverable right inline — files preview just like in chat. You can also get an email when a job finishes, fails, or needs you, with a link that jumps straight to it. There's a toggle to turn those emails off.
Automations: reusable, scheduled tasks
An automation is a saved task — a name plus a set of instructions — that can run over and over. Every run shows up as a job in Activity.
Create one
Open Automations (the calendar-clock icon in the left sidebar):
- In New automation, give it a Name (e.g. "Daily metrics digest") and Instructions (what it should do each run).
- Click Create automation.
- Click the automation to manage it — you'll see its run history and trigger settings.
Add a trigger
Inside an automation, choose how it should fire:
- Schedule — pick hourly / daily / weekly / monthly, a time, and a timezone. It shows you the next run time.
- Webhook — get a unique URL; anything that hits it starts a run.
- Email — get a unique inbound address; emailing it starts a run.
Each trigger can be enabled, disabled, regenerated (to rotate its secret URL/address), or deleted.
Example: a "Monday morning competitor scan" automation on a weekly schedule that researches your competitors and emails you a summary.
Fire-and-forget: the operator
Together, these make "set it and walk away" real. Silico's operator can carry a task to completion across many sessions — claiming the job, advancing it step by step, checkpointing its progress, and resuming after any interruption — so tasks measured in hundreds of steps finish on their own. You kick it off, get on with your day, and check Activity (or your email) for progress.
Guardrails
Background and unattended work always runs with brakes on:
- A spend ceiling on every job — it warns as it approaches the limit and pauses at it rather than running up a bill. Raise the cap to continue. As budget gets tight, it also shifts to cheaper models automatically.
- Stall detection — long runs periodically check "am I still on track?" and pause a stuck or looping job instead of burning money.
- Retry limits — a job that keeps failing stops and is marked failed (with a notification), rather than retrying forever.
- Approval on risky actions — anything irreversible or outward-facing pauses for your OK. See Approvals & autonomous mode.
You can also steer an in-flight job: send it a note of guidance and it applies at the next checkpoint — no need to stop and restart.