Oracle Cloud Function encounters 404 'resource not found' when calling Generative AI Agent endpoint

[Claire29]

Setting up AI agent integration with Oracle Functions

I have an Oracle Function that needs to communicate with a Generative AI Agent. My goal is to connect this through Oracle APEX so the agent can execute SQL queries against our database.

The problem occurs when my function tries to reach the agent’s chat endpoint. Instead of getting a proper response, I receive a 404 error saying the resource cannot be found.

Error details I’m seeing:

{
  "status": "success",
  "ai_response": "Error calling AI agent: {
    'target_service': 'generative_ai_agent_runtime',
    'status': 404,
    'code': '404',
    'opc-request-id': '7891A23BC45D67890123456789ABCDEF/12345678901234567890123456789012/ABCDEF1234567890ABCDEF1234567890',
    'message': 'Unknown resource 8ac15fb2-ca33-4de6-a068-bc612bb301bb',
    'operation_name': 'chat',
    'timestamp': '2025-06-11T10:42:18.445123+00:00',
    'client_version': 'Oracle-PythonSDK/2.154.1',
    'request_endpoint': 'POST https://agent-runtime.generativeai.eu-frankfurt-1.oci.oraclecloud.com/20240531/agentEndpoints/ocid1.genaiagentendpoint.oc1.eu-frankfurt-1.amaaaaaabumsjqaa5ezbdogwvx8nrjyfdulahguzno8hdmco4fwqrjg3ampr/actions/chat'
  }"
}

My implementation code:

import json
import oci
import uuid
from oci.auth import signers

def invoke_ai_assistant(user_input):
    try:
        # Initialize resource principal authentication
        resource_signer = signers.get_resource_principals_signer()
        
        ai_client = oci.generative_ai_agent_runtime.GenerativeAiAgentRuntimeClient(
            config={}, 
            signer=resource_signer
        )
        
        conversation_id = str(uuid.uuid4())
        
        response = ai_client.chat(
            agent_endpoint_id="ocid1.genaiagentendpoint.oc1.eu-frankfurt-1.amaaaaaabumsjqaa5ezbdogwvx8nrjyfdulahguzno8hdmco4fwqrjg3ampr",
            chat_details=oci.generative_ai_agent_runtime.models.ChatDetails(
                user_message=user_input,
                session_id=conversation_id
            )
        )
        
        return response.data
        
    except Exception as error:
        return f"Error calling AI agent: {str(error)}"

What I have verified:

• Both the AI agent and endpoint are created in eu-frankfurt-1 region
• Dynamic Group is properly configured for the Function
• IAM policies grant generative-ai-family access to the right compartment
• Resource principal authentication works without auth errors
• All OCIDs are copied directly from the Console
• Agent status shows as active
• Using OCI CLI 3.54.5 but gen-ai commands seem unavailable

Why might the agent endpoint return this ‘unknown resource’ error even though everything appears configured correctly?

[sophiac]

This looks like a resource lifecycle issue, not a config problem. That 404 ‘unknown resource’ error usually means the agent endpoint shows up in console but isn’t fully provisioned in runtime yet. I’ve hit this multiple times with OCI’s newer AI services. Check the endpoint’s actual state through REST API or wait a few more minutes if you just created it. There’s often a delay between console showing ‘active’ and runtime actually recognizing the endpoint. Also make sure you’re using the right service endpoint URL for eu-frankfurt-1 - I’ve seen the SDK use cached or wrong regional endpoints. Another thing - your compartment OCID in the endpoint resource might not match where your function’s trying to access it. Double-check the full resource path and make sure your dynamic group policies include the specific compartment where the endpoint lives.

[danielr]

The Problem:

You’re receiving a “404 Unknown Resource” error when trying to connect your Oracle Function to an OCI Generative AI Agent using the Oracle Python SDK. The error occurs despite the AI agent and endpoint appearing correctly configured in the OCI console. The function uses resource principal authentication and appears to have the correct IAM permissions.

Understanding the “Why” (The Root Cause):

This error, despite appearing to be an authorization issue, often indicates a problem with the lifecycle of your OCI Generative AI Agent or its endpoint. Even if the OCI console displays the agent and endpoint as “active,” there might be a delay before the runtime environment fully recognizes and provisions them. The service might also experience transient issues, leaving the endpoint temporarily unavailable even if it’s correctly configured. Additionally, there might be a mismatch between the compartment where the function is running and the compartment where the AI agent endpoint resides.

Step-by-Step Guide:

1. Verify AI Agent and Endpoint Existence and Status:

   • Go to the OCI console.
   • Navigate to AI Services > Generative AI Agents.
   • Verify that the AI agent linked to your endpoint (ocid1.genaiagentendpoint.oc1.eu-frankfurt-1.amaaaaaabumsjqaa5ezbdogwvx8nrjyfdulahguzno8hdmco4fwqrjg3ampr) actually exists and is in an active state. If not, recreate the agent.
   • If the agent exists, check the status of the endpoint itself. Sometimes, endpoints can show as “active” in the console but not be fully provisioned in the runtime environment.
   • Try recreating the endpoint.

2. Check Compartment Consistency:

   • Ensure that the Oracle Function and the Generative AI Agent endpoint reside in the same OCI compartment. A mismatch here will cause the function to be unable to access the endpoint, even with the correct OCIDs and IAM policies. Verify this in both the Function’s configuration and the Agent Endpoint’s details.

3. Inspect Function’s Network Configuration:

   • Make sure your Oracle Function’s Virtual Cloud Network (VCN) has the necessary rules to communicate with the Generative AI Agent service. Check for any firewalls or security lists that could be blocking outgoing requests. The request_endpoint in your error message shows the service endpoint used; ensure your network allows traffic to that URL.

4. Review IAM Policies:

   • Double-check that the IAM policies associated with the dynamic group used by your Oracle Function grant the necessary permissions to access the Generative AI Agent runtime service in the correct compartment. Pay close attention to the policy’s scope and make sure it covers the specific agent and endpoint OCIDs. Verify the policy includes actions like oci.generative_ai_agent_runtime.GenerativeAiAgentRuntimeClient.chat.

5. Test with a REST Client:

   • Use a REST client (like Postman) to send a test request directly to the agent endpoint using the URL from the error message and the same authentication method (resource principal). A successful test request here confirms the SDK isn’t the problem; if it fails, it helps isolate the issue to the endpoint itself.

6. Adjust Session ID and Timeout:

   • Consider using a simpler session ID format instead of uuid.uuid4(). Try a timestamp string instead to see if that resolves any session initialization problems.
   • Increase your Oracle Function’s timeout to at least 60 seconds, allowing sufficient time for the initial AI agent service warmup.

Common Pitfalls & What to Check Next:

• OCI Service Outages: Check the OCI service health status to rule out any planned or unplanned outages affecting the Generative AI Agent service.
• SDK Version: Ensure you’re using the latest version of the Oracle Cloud Infrastructure Python SDK compatible with the Generative AI Agent runtime service API. Version mismatches can lead to unexpected errors.
• Regional Endpoints: Manually specify the correct service endpoint URL for the eu-frankfurt-1 region in your oci.generative_ai_agent_runtime.GenerativeAiAgentRuntimeClient configuration to avoid relying on potentially cached or incorrect regional endpoint information from the SDK.

Still running into issues? Share your (sanitized) config files, the exact command you ran, and any other relevant details. The community is here to help!

[Oscar64]

Check your SDK version first - it needs to match the Generative AI Agent runtime service API. I hit the same problem when my Python SDK was out of sync. This service is pretty new, so version compatibility is way more important than with other OCI services. Update to the latest SDK or manually set the service endpoint URL in your client config instead of using SDK defaults. Also double-check your agent endpoint has the right knowledge bases and data sources linked. Sometimes the endpoint gets created but loses its agent connection during setup. Your error shows the request is reaching the service, but it can’t find that specific endpoint in the runtime registry. I’d test the endpoint directly with a REST client using the same auth first - that’ll tell you if it’s an SDK problem or if the resource is actually missing.