Scheduled routines

A routine is a scheduled agent workflow. You give it a prompt, choose the skills it may use, pick a context mode, and set a schedule. From then on Forven runs that prompt on its own cadence and records each run as a task in the queue.

This page is for users who want recurring agent work — a morning research sweep, a periodic data check, a scheduled cleanup — without launching it by hand each time. You create and manage routines on the /routines page.

Routines are part of the agent layer, so they require the Forged tier. The full agent layer is unlocked for everyone during beta.

Routines versus the scheduler

Forven has two scheduling systems, and it helps to keep them apart.

• The scheduler runs Forven's own machinery: the 35+ built-in jobs that drive the pipeline — the promotion loop, the verdict loop, the live scanner, maintenance, and so on. You tune those from /ops, not from /routines. See Scheduler & jobs.
• Routines are your scheduled prompts. They are agent workflows you author: a prompt plus skills plus a schedule. They run on top of the system, not as part of its core loop.

If you want to change how often Forven optimises strategies, that is a scheduler job. If you want an agent to summarise overnight research every weekday at 8am, that is a routine.

Anatomy of a routine

Every routine is defined by a small set of fields.

Context modes

The context mode tells the agent what kind of run this is and how much surrounding state it should carry. Forven exposes four:

• scheduled — the default for unattended, recurring work. Use this for routines that run on their own with no operator watching.
• interactive — for routines you expect to sit with and steer.
• recovery — for runs aimed at cleaning up or recovering from a previous failure.
• research — for research-oriented sweeps.

Pick the mode that matches the intent of the routine; scheduled is the right choice for most recurring jobs.

Cron and presets

A routine's schedule is a standard cron expression. If you would rather not write one, the create form offers presets:

• Hourly
• Daily
• Weekdays
• Weekly
• Monthly

Whatever you pick — preset or hand-written cron — the form shows a live preview of the next run times so you can confirm the cadence before you save.

Creating a routine

You build routines on the /routines page.

Steps

1. Open Routines from the sidebar to reach /routines.
2. Click Create Routine.
3. Enter a name and the prompt — what you want the agent to do on each run.
4. Add skills from the autocomplete input. These bound what the agent may use; only add what the prompt needs.
5. Choose a context: scheduled, interactive, recovery, or research. Use scheduled for unattended jobs.
6. Set the schedule — write a cron expression, or pick a preset (Hourly, Daily, Weekdays, Weekly, Monthly).
7. Check the next run times in the live preview to confirm the cadence is what you intended.
8. Click Create. The routine appears in the list.
9. Toggle enabled to start scheduling it. Use Run now if you want it to execute once immediately.

What you'll see

The /routines page is a list of routine rows with status badges. The create form has a cron picker with the presets above and the next-run preview. Each row carries inline edit, run now, pause/resume, and delete controls, so you can manage a routine without leaving the list.

Managing routines

Once a routine exists you can:

• Edit it inline to change the prompt, skills, context, or schedule.
• Pause / resume it by toggling enabled — a paused routine keeps its definition but stops firing.
• Run now to trigger an immediate run outside the schedule, which is useful for testing a prompt.
• Delete it when you no longer need it.

Watching a routine run

A routine doesn't show its work on the /routines page — it shows it in the task queue. Each time a routine fires, it creates a task with the usual lifecycle:

pending → running → (done | failed | blocked | rejected)

To follow a run, open /tasks and find the task it created. The task's detail page (/tasks/[id]) gives you the full audit log, the tool calls the agent made, and any error context — the same inspection surface you'd use for any other agent work. If a scheduled run fails, that's where you read why.

Caveats

• Forged tier. Routines are an agent-layer feature, so they belong to the Forged tier. Beta unlocks the agent layer for everyone.
• A routine is a prompt, not a guarantee. It instructs an agent; the agent does the work within the skills and context you granted, and the outcome lands in the task queue like any other run. Review failed runs in /tasks.
• Routines aren't the system scheduler. Disabling or deleting a routine has no effect on Forven's built-in pipeline jobs. To pause those, use system controls and the scheduler on /ops.
• Agents never place trades directly. A routine cannot bypass Forven's safety model. Capital-affecting actions still flow through the same gates and approvals as everything else, and beta builds are hard-locked to paper regardless.

Forven is a research tool. Anything a routine produces is part of a research process — results are not predictive, and nothing here is financial advice.

Related

• Scheduler & jobs — Forven's built-in pipeline jobs and their cadence.
• Tasks & the queue — where each routine run is recorded and inspected.
• Agents — the agent layer that routines drive.