Runs & schedules
A run is one occasion of an agent doing its job. A schedule is what makes runs happen without you asking.
This is the step that separates an agent from a chatbot. A chatbot works when you type. An agent can work at 7am on Monday whether or not you are awake.
What starts a run
Four things:
| Trigger | What it means |
|---|---|
| Chat | You talk to the agent and it answers. The run you can watch happen. |
| Schedule | A clock fires — "every weekday at 07:00". |
| Flow | Something happened elsewhere in Nextcloud and triggered the agent. |
| Webhook | An outside system called in. |
The agent does the same work regardless of which one started it. The difference is only who pressed the button — and whether anyone is watching.
What happens during a run
- The agent assembles what it knows: its prompt, its skills, its context, and its memory.
- If RAG is on, it looks up material relevant to the task.
- It thinks, and may call tools to take action.
- If it reaches something gated, it pauses for an approval and waits for a human.
- It delivers its answer.
Every run is written to the audit trail — what it did, which tools it called, what it cost. That record is what makes an unattended agent something you can account for afterwards rather than something you hope went well.
Schedules
A schedule attaches to an agent and fires it on a rhythm — a cron expression, so anything from "every hour" to "the first Monday of the month" is expressible.
Scheduled runs are where agents earn their keep: a morning briefing, an overnight inbox triage, a weekly report. Nobody has to remember.
Because a scheduled run has no human watching it, two things matter more than usual:
- Delivery. Output goes to a Nextcloud Talk conversation, so the work lands where your team already is — falling back to a notification if Talk is not installed. An unattended run that writes into the void is useless.
- Gates. A schedule can require an approval before it dispatches, and an organisation-wide kill-switch can stop scheduled work entirely. If an agent is going to act while you sleep, you want a brake.
Cost
Every run costs something — cloud models bill per token. Hermiq records token usage and cost per run, and an administrator can set budgets and cost guardrails per organisation, so a misbehaving schedule cannot quietly run up a bill. Usage is visible on the dashboard.
This is a real consideration for scheduled agents specifically: a chat run costs what one conversation costs, but an hourly schedule costs that every hour, forever, whether or not anyone reads the output.
Where to find it
The Dashboard shows run metrics — how many, success rate, latency, tokens. An agent's own runs and their history are on its detail page. Approvals holds anything waiting on a human.