Skip to content

Pause and error codes

A paused run shows its code in Error Code. After you fix the cause, rescue the run. See Fix a stuck run.

CodeMeansDo this
BUDGET_EXCEEDEDThe run used up Max Model Calls, Max Tool Calls, or Max Total Tokens.Raise the limits Operate names on the agent, then rescue.
MODEL_OUTPUT_TRUNCATEDThe model’s answer hit Max Output Tokens.Raise Max Output Tokens on the Agent Model, then rescue.
MODEL_REFUSEDThe model refused to answer.Read the run story. Rescue with guidance, or cancel.
DISPATCH_UNDELIVERABLECore tried three times, waiting longer each time, to start the next step as the execution user and could not.Check that the execution user is active and that Verify passes on its identity. Make sure whoever continues runs can use the Named Credential. Then rescue.
UNKNOWN_UNRESOLVEDNobody reconciled an Unknown call before the run’s payload retention ended.Reconcile the call, or cancel the run. See Settle an Unknown call.
WAIT_EXPIREDAn outside wait, or an approval set to Pause for Operator, ran out of time.Rescue to let the agent continue without it, or cancel.
TOOL_BINDING_CHANGEDA tool changed or was deactivated after the call started.Restore the tool as it was, or cancel the run and start a new one.
TOOL_AUTHORIZATION_POLICY_INVALIDThe agent’s policy asked for approval with no reason or a deadline out of range.Fix the policy class, then rescue.
APPROVAL_ASSIGNMENT_UNRESOLVEDNobody eligible could be found to approve the call.Set Approver Type, Approver User, or Approver Queue on the tool, then rescue.
INITIATING_USER_APPROVAL_UNSUPPORTEDA call needed approval, but the agent has no execution identity.Give the agent an identity, or change the policy. Then start a new run.

These do not stop the run. The agent reads them and decides what to do next.

CodeMeans
TOOL_APPROVAL_DENIEDThe approver rejected the call. The agent gets the approver’s comment.
TOOL_APPROVAL_EXPIREDNobody decided in time, and the tool is set to Deny and Continue.
TOOL_AUTHORIZATION_DENIEDThe agent’s policy denied the call. The agent gets the policy’s reason.
TOOL_NOT_RUNCore did not run the call, for example the rest of a batch after one call failed.
TOOL_BATCH_NOT_ALLOWEDThe model asked for several tools at once, but the agent allows one per turn. Core asks again.

A tool’s own errors, such as a validation rule message, also go back to the agent with the tool’s code.

Pluto is licensed under the GNU Affero General Public License, version 3 only.