When setting up a Conversational Agent in Agent Builder (setting system instructions, model selection, tools, etc.), clicking Save triggers a PATCH request that returns a 500 Internal Server Error. The agent was created successfully, but no configuration changes can be persisted. I get the error “Something went wrong. Please try again later.”
Looks like a backend-side failure in Vertex AI Agent Builder, not a misconfiguration on your end. Your troubleshooting already covers most common causes.It’s likely not your fault if:
Error is 500 (not 403/400)
Happens across browsers
Minimal configs also fail
API calls fail too
Open a Support Ticket with:
Project number
Agent ID
Region (us-west1)
Exact failing endpoint
Tell them or mention that:- Consistent 500 INTERNAL error on PATCH for lowcodeAgents - reproducible across configs and browsers. Likely backend validation or regional support issue.
I have exactly the same problem. I tried everything like the other reporter. All the AI help suggestions have been implemented and still cannot get and agent saving. Very frustrating.
I’m experiencing the same issue! I noticed this in my API/Service logs. I’ve also tried using two different accounts, but neither of them saved the data.
I got it solved using the following list of commands in the cloud shell.
Here is the consolidated list of commands to fix the 100% error rate on the UpdateLowcodeAgent method for your project (PII Removed by Staff).
Step 1: Fix IAM Permissions (Run in Cloud Shell)
This grants the specific “Reasoning Engine” role to the Google-managed service account that handles low-code updates.
# Set variables for convenience
PROJECT_ID="xxxxxxxxxxxx"
PROJECT_NUMBER=$(gcloud projects describe $PROJECT_ID --format='get(projectNumber)')
# Grant the mandatory Reasoning Engine role
gcloud projects add-iam-policy-binding $PROJECT_ID \
--member="serviceAccount:service-$PROJECT_NUMBER@gcp-sa-aiplatform-re.iam.gserviceaccount.com" \
--role="roles/aiplatform.reasoningEngineServiceAgent"
# (Optional) Grant storage access if your agent uses Cloud Storage buckets
gcloud projects add-iam-policy-binding $PROJECT_ID \
--member="serviceAccount:service-$PROJECT_NUMBER@gcp-sa-aiplatform-re.iam.gserviceaccount.com" \
--role="roles/storage.objectAdmin"
Step 2: Update the Vertex AI SDK (Run in Cloud Shell)
Ensure your environment has the latest library version to support the LowcodeAgentService.
pip install --upgrade google-cloud-aiplatform
Step 3: Run the Update Script (In Python)
To execute the update, start the Python interpreter and run the initialisation.
Enter Python:
python3
Paste and Run these commands:
from google.cloud import aiplatform
# Initialize with your specific project and region
aiplatform.init(project (PII Removed by Staff))
# Note: Replace 'YOUR_AGENT_ID' with the actual ID from your console
# agent = aiplatform.ReasoningEngine(PII Removed by Staff)
# agent.update(display_name="Updated Agent Name")
Essentially I did an update and then ensured that a service had the right permission.
This service had the permission shown in the interface but they were not there when tested in the shell
Hi Chris! A 500 Internal Server Error when saving an Agent Builder configuration is usually caused by regional mismatches or hidden quotas. Here are a few things you can check:
Data Store Region Mismatch: If your Agent is located in us-west1, but you attached a Data Store or Tool that is located in global or us-central1, Google Cloud will silently fail to link them and throw a 500 error.
Model Availability: Check if the specific LLM you selected for the agent (like gemini-1.5-pro) is fully supported for Agent Builder in us-west1. Try creating a test agent in global or us-central1 to see if it saves successfully.
Logs Explorer: Sometimes the UI hides the real error. Go to Cloud Logging > Logs Explorer and filter by resource.type="aiplatform.googleapis.com" or dialogflow to see if there is a more descriptive backend error causing the 500 block.
Let me know if deploying it to global fixes the save issue!"
