An agent that runs on meetings you weren't on, fires once for every record in a bulk change, or uses more credits than you planned usually comes down to three things: what starts it, what it can see, and how often that happens. Every run uses credits, including a run where the agent decides there is nothing to do, so the fix is to stop unwanted runs from starting rather than ask the agent to skip them.
New to agents? Start with Agents. For writing the trigger and instructions themselves, see How to build a great agent.
Why an agent runs when you didn't expect it
A trigger watches the whole workspace. A trigger fires for every matching record or meeting in the workspace, not only the ones you created or attended. Filters are what narrow it.
An agent runs as its owner and only sees what they can open. An agent that watches meeting recordings runs only on meetings its owner can open: meetings they were on, plus any shared with them. If your workspace shares meetings with everyone, that is every recorded meeting.
Private does not limit what an agent runs on. Setting an agent to private or public does not change which records or meetings start a run.
Each copy of an agent runs separately. If you and a teammate both have the same meeting agent and you were both on a call, both copies run on it and each run uses credits.
New agents start switched on. Agents Rep builds for you are on as soon as they are created, and agents picked during setup are on as soon as someone clicks Enable.
One change to many records means many runs. A trigger fires once per matching record, so updating a field on 60 records, by hand or by asking Rep, starts up to 60 runs of an agent that watches that field.
Narrow when it runs
Go to Agents and open the agent.
Under Triggers, click a trigger to open it, then click Add filter so it only fires for the records you want. For example, a Where condition of Stage Changed to one stage, instead of any change to a deal.
Check the estimate on the trigger, such as Triggered 2 times, past week. It counts matching events from the past week, so you can see the volume before the agent runs on it.
For a rule a filter can't express, such as "only meetings I attended" or "only external meetings", add it as the first line of the instructions. The run still starts and uses a small amount of credits before the agent stops, so use a filter wherever one fits.
See How to build a great agent for filter patterns and how to write the instruction gate.
See how often it runs and what it costs
On the agent: the agent's page shows its average credits per run next to the owner's name. Click Insights at the top of the page for more. The Usage card shows Total runs, Success rate and Avg cost per run for the last 7 days, last month or last year.
Each run: the agent's activity log lists every run and what triggered it. Cost is shown as an average, not per run.
Across the workspace: go to Settings → Workspace → Billing. On the Credit usage card, By feature lists each agent as its own row under agents. By member, where available, counts an agent's spend against the person who owns it.
Stop an agent that is spending
Turn it off. Open the agent's ⋯ menu and choose Disable agent. Any workspace admin can do this for any agent, including ones they did not create.
Stop a run in progress. A run that is still going shows a Stop button in its activity view.
Leave unwanted paused runs alone. Runs that reach the hourly limit or your credit limit pause instead of failing, and the agent shows a Paused runs banner. Paused runs use no credits while they wait. Resume starts all of them and uses credits for each, so only resume runs you actually want.
Use approval while you test
Under Tools and access on the agent, set each tool that sends email, changes records or posts to Slack to Ask permission instead of Always allow. The agent pauses and asks you before that action. This controls what the agent does, not whether it runs, so it does not reduce how often the agent starts.
Before you turn an agent on
Check Clarify doesn't already do it. Some work is automatic and free, and an agent repeating it pays for every run. See Agents → Keep agent costs down.
Pick the narrowest trigger and add filters. Check the trigger's past-week estimate, and once the agent has run, the weekly credit estimate under Triggers.
Set every tool that writes to Ask permission.
Click Run agent to test it once, then read the run in the activity log.
After the first day, open Insights and check Total runs and Avg cost per run before leaving it on.
If teammates run their own copy of the same meeting agent, decide whether one shared agent would do.
FAQ
Why did my agent run on a teammate's meeting?
Because you, as the agent's owner, can open that meeting. It was either shared with you or your workspace shares meetings with everyone. To limit the agent to meetings you attended, add that rule to its instructions.
Why didn't my agent run on a meeting?
The agent's owner can't open it. An agent only runs on recorded meetings its owner can see. To include a teammate's meetings, the teammate can share them with the owner. See Email and meetings.
Does a run cost credits if the agent decides to do nothing?
Yes. Every run that starts uses at least one credit. A filter on the trigger is the only way to avoid the run.
Can an admin stop a teammate's agent?
Yes. Any workspace admin can turn any agent off from its ⋯ menu.
Where do I see which agent is using my credits?
In Settings → Workspace → Billing, on the Credit usage card, under By feature. For what else uses credits, see What uses credits, and how to control the spend.