Steer a running agent workflow

September 28, 2026

A run is heading to the wrong specialist. You can see which node is working and which edge it is about to take. In LiveGraph you drag that edge, and the next hop follows. The model call already in flight finishes and stays on the trail. Nothing is restarted.

That is possible because the engine never plans the whole run. It dispatches one hop, then reads the graph again and decides the next hop from what the graph looks like at that moment. A drag during a run you are watching is stored on that run. The saved graph, and therefore the next scheduled run, changes only if you keep the route.

Who this is for

A founder or a small agency that already runs agents against a real system — a repo, an inbox, a site — and is on the hook when a hop goes to the wrong place. Schedules and webhooks stay pinned: they snapshot the graph at the start, so an unattended job ignores a drag. Chat and the canvas Run button start live, which is the run a person is watching.

The platform price is flat. Solo is free. Team is one monthly price per workspace. Model spend is your keys at provider prices, or our hosted keys metered on top — either way, LiveGraph does not meter runs.

Try the drag

The live demo needs no account and no API key. A scripted model holds a hop open long enough to drag the edge and watch the next hop take the new target. When you want your own graphs, sign up and the included model runs them. Bring a key when you want a different provider.

If you are leaving a canvas that freezes when you press run — Flowise is unmaintained, and OpenAI Agent Builder shuts down on November 30, 2026 — start with the demo, then the Flowise importer or the Agent Builder migration guide.