A workflow is a process you draw instead of describe. In the visual editor you drag steps from the node library onto a canvas, connect them in order, and the result is a pipeline the agent runs the same way every time — on demand when you trigger it, or on a schedule while you are doing something else.
Building one
The editor has two parts: a node library and a canvas. The library is a flat list of every available step — find the node you want and drag it onto the canvas. Connecting two nodes sets the order they run in, with the output of one step feeding the next.
A sensible rhythm for a first workflow:
- Drag the steps you need onto the canvas, roughly in order.
- Connect them so the sequence reads without ambiguity.
- Run it once on demand, then read the step results in Activity.
- Only once a manual run comes back clean, put it on a schedule.
The last habit matters. A schedule multiplies whatever behaviour a workflow already has, good or bad — prove the flow by hand first.
On demand or on a schedule
Every workflow can run both ways. On-demand runs suit processes you trigger deliberately: a release checklist, a data pull you do when asked. Scheduled runs suit genuinely recurring work — the same steps, fired at the interval you set, without anyone pressing the button.
A worked example
Say you want a Monday-morning competitor summary. The canvas might read:
[Fetch competitor updates] → [Summarise findings] → [Format as a brief] → [Deliver]Four nodes, three connections. Run it by hand on a Friday to check the brief reads well, then schedule it for Mondays at 08:00. From then on, each Monday's run appears in Activity, and opening it shows exactly what every step fetched, wrote, and produced.
Cost and models
A workflow run is agent work, and it is metered like any other agent work: the run draws credits based on the compute and model time its steps actually use. A short flow costs little; a heavy one costs what its steps cost. There is no separate workflow meter.
Model choice is automatic. You never pick a model per node — Kaer routes each step across its stack of frontier models based on what the step needs. See Billing and Usage for how credits work in general.
When a run fails
A failed run is logged in Activity like any other, with the step-by-step trail intact, so you can see exactly which node failed and what it received. Fix the step or its input, run once more on demand, and only then let the schedule carry it again.
Good to know
- Do scheduled runs cost credits too? Yes — every run is metered the same, whether you triggered it or the schedule did.
- Where is the history? Every run, with full step results, is in Activity. Filter by type to see workflow runs only.
- When is a workflow the wrong tool? When the work is one-off or still taking shape. Run it as a normal agent turn first, and promote it to a workflow once the steps have stabilised.
- Can someone else follow my workflow? That is half the point — the canvas is the documentation. A colleague can read the nodes left to right and understand the process without reading any prompt text.