Recurring jobs
Scheduled AI agents and recurring jobs
A scheduled AI agent is a request that runs by itself at the time you pick, every weekday, every Monday or on the first of the month. In Evriq a recurring job lives on a project, runs as the person who set it up, and leaves its results where the team already works.
Start a session or post a message on a schedule
A recurring job does one of two things each time it fires. It starts a fresh session with instructions for the agents, or it posts a message into a session that already exists, so a standing conversation picks up where it left off.
You add a job from the project's Recurring jobs section. Give it a name, say what the agents should do and what a good result looks like, then pick when it runs. Any member can schedule jobs on the projects they can see. Administrators can also schedule jobs that run commands on the server.
A job runs as the person who added it and counts against that person's limits. If its owner leaves the project or the workspace, the job stops running. A job set to run once runs one time and then pauses itself.
Pick when it runs, in your time zone
The schedule picker offers every weekday, every week on a chosen day, every month on a chosen date, or a custom schedule. It starts in your browser's time zone and you can change it, so a report set for 7:10 in the morning arrives at 7:10 where your team is.
The picker shows the next times the job will fire before you save it. The full cron expression is there under Advanced for anyone who wants it.
- Every weekday, for a morning summary of what changed overnight
- Every week on a day, for a status report or a pipeline review
- Every month on a date, for a close checklist or a renewal sweep
- A custom schedule, for anything else
Credentials for jobs that run unattended
A job that reads a mailbox, calls an API or queries a database needs a credential at the moment it runs. You save that credential as a secret, for the whole workspace or for one project, and it reaches every job run it applies to as an environment variable of its own name. Nobody has to be there to open a vault first.
Secret values are write-only once saved, stored encrypted, and masked out of job output on a best-effort basis. Every change and every read into a run is recorded by name, never by value.
When a job meets a check only a person can pass
A job that drives a website can meet a captcha or a page that asks you to prove you are human. The job never tries again and never works around it. It stops, and the check appears in your inbox and as an alert.
Open it and you see the page the job is waiting on as a live picture. Click and type as you would on the site, press Done when you have solved it, and the job carries on from that same page. Every look and every action is in the audit log, without what was on the page or what you typed.
See every run
Show its runs lists every run of a job and whether it succeeded. A run that is still going can be stopped by the person who made the job.
When a session job's agent writes an artifact, it is filed on the project with the run's date after its title, so a weekly report becomes a dated series on the board. A failed run shows on the project's activity with its reason and lands in its owner's inbox, so a silent failure is not something you find out about a week later.
Sessions opened by a job that succeeded with nobody writing in them are retired soon after, which keeps the open list short. Nothing is deleted, and a retired session keeps its full transcript. If you want people to check a job's output before anyone acts on it, add a review round to the instructions.
Questions about scheduled AI agents
How do I schedule an AI agent to run every week?
Open the project, add a recurring job, describe the work, and choose every week on the day you want with the time and time zone. The jobs help page walks through each field.
Whose account does a scheduled job run on?
A job runs as the person who added it. Its turns count against that person's limits and run on your workspace's own provider keys.
What happens when a scheduled job fails?
The failed run appears on the project's activity with its reason and in the job owner's inbox. From there you can open the run, read the log and decide what to do next.
Can a job use a password or API key?
Yes. Save it as a workspace or project secret and it arrives in each run as an environment variable, with nobody needing to be there to open a vault first.
Can I stop a run that is already going?
Yes. The job's creator can stop a run in progress from its list of runs, and an administrator's Stop all turns stops every running agent in the workspace at once.