Scheduled jobs

Scheduled jobs let SovrGPT carry out a task regularly without you, for example "Every weekday at 9 a.m., summarise the news on funding programmes". The result lands in a chat and, if you like, in your inbox.

Only you can see, change and run your jobs, in a team organisation too. Owners and admins see only an overview: who runs which job on which schedule and how the last run went. They do not see the prompt, the chat or the results, just as with your chats.

Creating a job

1. Open Geplante Aufgaben (scheduled jobs) in the sidebar and choose "+ Neue Aufgabe" (new job).
2. Enter a title and a prompt (up to 8,192 characters).
3. Set the cron expression and time zone (default Europe/Berlin).
4. Optionally choose a model and an agent; the agent brings its system prompt along.
5. Choose tools and delivery, then save.

Web search: it is preselected when you create a job and sends the model's search queries to Brave Search, a provider based in the USA. The same switch also lets the model fetch public web pages. If the job needs nothing from the web, untick it. Other tools and MCP connections are not available here, because nobody is there to approve an action.

Model: without a choice, the job runs on the self-hosted default model; after a pause the result then arrives a few minutes later. For punctual results, pick a model without a cold start.

Cron syntax

The usual five fields, evaluated in the job's time zone:

SovrGPT checks every five minutes what is due. A run therefore usually starts up to five minutes after the scheduled time.

Where does the answer end up?

In the job's chat: the first run creates it, and every later run appends prompt and answer. Tick "Ergebnis zusätzlich per E-Mail an mich" (also email me the result) to get the answer by email as well, with a link to the chat. Each run sees only the prompt, not earlier answers; answers are capped at about 1,500 tokens.

Guarantees & limits

• At most one run per slot, even if several checks run at the same time.
• No automatic retry: the overview shows the error, and the next run follows at the next scheduled time. "Jetzt ausführen" (run now) starts an extra run.
• Spending limits apply here too; if one is reached, the run stops with a reason.

API for jobs

There is no REST API for this; you manage jobs only in the UI under /orgs/<slug>/scheduled. If you need one, let us know. The existing endpoints are listed in the API reference.