Kill-switch
Stop one agent or every agent on a project instantly. The next statement is refused, with no credential rotation.
The kill-switch stops an agent now. Trip it on a single credential to cut off one agent, or at the project level to stop every agent at once. The next statement on a killed credential is refused. There is no credential rotation and no change to your database.
Trip the kill-switch
# Stop one agent (reversible; re-enable with `pgbeam agents enable`)
pgbeam agents disable agt_xxx
# Stop every agent on the project
pgbeam projects update --agents-disabled trueOn the Credentials tab, select a credential and disable it. To stop every agent at once, engage the project-level kill-switch in the project settings.
# Stop one agent: set its credential status to disabled
curl -X PATCH https://api.pgbeam.com/v1/projects/{projectId}/agents/{agentId} \
-H "Authorization: Bearer pbo_..." \
-H "Content-Type: application/json" \
-d '{"status": "disabled"}'
# Stop every agent on the project: engage the project kill-switch
curl -X PATCH https://api.pgbeam.com/v1/projects/{projectId} \
-H "Authorization: Bearer pbo_..." \
-H "Content-Type: application/json" \
-d '{"agents_disabled": true}'Reverse either one by sending {"status": "active"} to the credential or
{"agents_disabled": false} to the project.
How it takes effect
The kill state streams to the data planes and applies on the next statement. A
killed agent gets a clear error over a connection string, and a 401 over the
hosted MCP endpoint.
Kill-switch vs revoke
| Action | Effect | Reversible |
|---|---|---|
| Kill-switch | Pauses access immediately. Credential and policy are preserved. | Yes |
| Revoke | Removes the credential entirely. | No |
Use the kill-switch when you want to stop an agent now and decide later. Use revoke when the credential should be gone for good.
Org-level safety net
When an organization-level budget ceiling is breached, the control plane can trip the kill-switch automatically. See Budgets.
Related
Query Budgets
Cap queries per window and rows per result for an agent credential. Runaway loops and full-table scans hit a ceiling instead of your database.
Honeytokens
Register decoy relations that no legitimate agent should ever touch. An agent query that references one is blocked and recorded as a canary_tripped event, with an optional automatic kill.