Test an automation before it runs for real, then run it and follow what it does. A monthly budget cap keeps its AI cost within a limit you set.
Before you begin
- Testing and running need the Manage automations permission.
- Run now needs the automation to be active. See Build an automation.
Test a single step
-
Select the step
Open the automation in the builder and select the step you want to check.
-
Run its test
Under Test this step, choose Run test. Choose Re-test to run it again.
-
Read the result
A read step shows "Ran read-only. Output pinned for the next step." The next step's test uses that pinned output. A send or approval step shows "Nothing was sent, preview only". Choose Clear pin to drop the pinned output.
Test the whole flow
Choose Test in the command bar. Hoppflow runs the whole flow as a dry run and sends nothing. The result reads one of:
| Result | Meaning |
|---|---|
| "Ran end-to-end, nothing sent" | The flow passed. You can activate it. |
| "Stopped at a failing step" | Open the failing step, fix it, and test again. |
| "Can't test yet" | The flow is not ready to test, for example because it has no steps yet. |
A passing whole-flow test is required before Activate works.
Run an automation
-
Activate it
Choose Activate after a passing test.
-
Start a run
Choose Run now on the expanded automation card or in the builder. If Hoppflow asks "Run this automation?", choose Run it.
-
Approve the paused steps
In the builder, Hoppflow confirms "Run started. Side-effecting steps wait in your approvals inbox." On the card it says "Run started. Results appear below shortly." Open Inbox and review each step that waits for approval.
-
Let the run continue
After you approve, the run continues with the values it had before the pause, so what goes out is what you approved.
Limits on a single run
- A run can make at most 100 sends.
- An approval left unanswered for 7 days ends the run.
- Repeated failed sends to a vendor can trip a workspace-wide circuit breaker that stops further sends for a while.
- A partly failed batch can leave earlier items sent and later items unsent. Check the run before you send again.
How sends from a loop are approved
A write step inside a loop that is set to Send once per item collects its messages into one batch approval. A batch holds at most 10 sends. Items past that are dropped, and the card notes how many the loop had. Looping over writes to connected apps is not allowed.
Set a monthly budget cap
AI steps in an automation use the workspace's pooled Credits. When a run starts, it reserves its estimated Credits. A run that cannot reserve them does not start. A run that runs out of Credits stops with "The workspace ran out of Credits, so the run stopped. The owner can add Credits in Billing, then run it again."
By default nothing is charged automatically. A workspace owner can opt in to Auto-refill to top up Credits.
-
Open the cost section
Expand the automation's card. In Cost, choose Set budget.
-
Enter the cap
Enter a Monthly budget cap (USD) and choose Save.
-
Check the chip
The card shows "Cap $X/mo", or "Over cap $X/mo" once the automation has spent past it.
The cap is the AI cost of the automation in US dollars per month. Once it is over the cap, new runs are blocked with "The automation budget was reached. Review its budget before running it again." Runs that resume after an approval still finish.
Follow a run
The automation card lists its latest runs under Runs, with "No runs yet." when it has none. Open a run to see how long it ran, its cost, the version it ran ("ran vN") and its status.
| Status | Meaning |
|---|---|
| Running | The run is working. |
| Waiting for approval | A step waits in Inbox. The run page shows "This run is paused for your approval" with Review in your Inbox. |
| Succeeded | The run finished. |
| Failed | A step failed and the run ended. |
| Cancelled | The run was cancelled before it finished. |
| Stopped, over budget | The run reached the automation's budget cap. |
The Agents page also groups runs under Needs you, Working and Recent. Choose Ask about this run to ask a question grounded in that run's session and your company brain. Your replies there do not steer the running agent, and its external actions still wait for your approval.
Explain why this automation did not finish. Find the first step that failed or paused, then check whether it was waiting on an approval, a connection that is not working, Credits, or a plan limit.Tell me the cause, which items it affected, and the safest next action. If the records do not show the cause, say so rather than guessing at one.