Troubleshoot Gemini Enterprise Agent Platform

This page is the starting point for troubleshooting Gemini Enterprise Agent Platform. Use it to diagnose an error, look up a specific error, find the guide for your task, or get support.

Diagnose an error

Before you look for a specific fix, collect the following information:

Read the error

Note the status code, such as 400 or 429, and the exact message text. You can look up both in the error catalog.

You can also identify the specific failure from a machine-readable reason, documentation link, or prefiltered log view if those are present in the error message.

Find the logs

To search and filter agent error logs, use the Logs Explorer, set Resource type to Vertex AI Reasoning Engine, and select the corresponding Resource container value (your project number) and Reasoning engine ID value.

Check quotas and service health

Errors that appear intermittently, or that started without a change on your side, are often caused by quota or by an ongoing incident:

Error catalog

Find the error you received in the following table. If your error isn't listed, use the troubleshooting guides for the task you were trying to complete.

Status code or message Resolution
ImportError: cannot import name 'reasoning_engines', ImportError: cannot import name 'agent_engines' Outdated version of the Agent Platform SDK for Python
401 Context-Aware Access requirements are not met, 401 Request had invalid authentication credentials 401 authorization errors
ValueError: Cannot get the Candidate text Content generation errors
failed to start and cannot serve traffic, on your first custom container (BYOC) deployment in a region BYOC first-time creation error
failed to start and cannot serve traffic or Request is prohibited by organization's policy, inside a VPC Service Controls perimeter VPC-SC violation errors
429, RESOURCE_EXHAUSTED Resource exhausted or rate limit errors
500 internal server error from a prebuilt template Prebuilt template errors
cloudpickle or pydantic version conflicts when serializing your agent Serialization errors
NotFound: 404 Can not copy from "gs://..." Cloud Storage bucket subdirectory isn't created
iam.serviceAccounts.actAs denied, metadata server unavailable Custom service account errors
InvalidArgument: 400 Provided filter is not valid Error when filtering the list of agents
google.api_core.exceptions.Unknown: None, Failed to convert project number to project ID, during startup Errors during Agent Runtime startup
403 Forbidden on an outbound request 403 Forbidden errors
Certificate verification failures reaching a self-signed or private CA destination Self-signed or private CA destinations fail to connect
TLS handshake failures from a custom container (BYOC) agent Agent deployed with custom container (BYOC) fails to connect
Sandbox creation failures Sandbox creation issues
Code execution times out, file I/O failures in a sandbox Code execution issues
RuntimeError: Failed to generate memory, or no memories returned after a session No memories were generated

Troubleshooting guides

If you don't have a specific error message, or your error isn't in the error catalog, see the following table for the corresponding guide for your task:

Task Guide
Installing the Agent Platform SDK and setting up your environment Agent Runtime environment setup
Creating an agent Create an agent
Deploying an agent to Agent Runtime Agent deployment
Calling an external service from an agent Agent Gateway connectivity
Listing, updating, or deleting a deployed agent Manage deployed agents
Running code in a sandbox Code Execution
Generating or retrieving memories Agent Platform Memory Bank
Serving a model or hitting a model quota Troubleshooting machine learning services

Get support

If the guides on this page don't resolve your issue, see Getting help for agents for support packages, community forums, and bug reporting. For billing questions, see Billing questions.

If you file a bug, provide the following information:

  • Status code
  • Error message
  • (Optional) Feedback if the error message is wrong, misleading, or not actionable.