Full automation is the wrong goal for plenty of processes. A refund over a threshold, a contract clause exception, a vendor onboarding — these want a human decision embedded inside an automated flow, not a fully manual process or a fully blind one.
FluidGrids supports this with run control as a first-class API. A run can be paused at the point where human judgment is needed. The run holds its full state — every completed node's output intact — while the decision happens on human time: minutes or days, it doesn't matter to the worker fleet.
The decision flows back in through resume-with-payload: when the approver acts, the run resumes and the payload — approved, rejected, an adjusted amount, a note — becomes data the downstream nodes branch on. An if node routes the approved path to fulfillment and the rejected path to a notification, exactly like any other data in the graph. Cancel remains available when the right decision is "don't continue at all".
Surrounding the gate, the usual machinery does the boring parts: a webhook trigger starts the run when the request arrives, connector nodes post the approval request to Slack or email with the context the approver needs, and the run detail shows exactly where each pending request sits — no more "whose inbox is this stuck in?".
Because pause, resume, and cancel are permission-guarded operations, you control who may release a gate. The audit story comes free: the run history shows the workflow version, the pause, the resume payload, and every node that acted on it — an approval trail that reconstructs itself.
Do it yourself
Embed an approval gate inside an automated run — pause at the decision point, let a human decide, and resume the same run with their answer as data the downstream nodes branch on.
In the builder, add an internal Wait node at the step where human judgment is needed.
You should see: The workflow can pause and hold its full state at that point.
Ready to make this your story?
