Troubleshooting a Persistent 404 NOT_FOUND Error with VertexAiSearchTool

[this-dave]

Hi everyone,

We're currently developing an agent using the Google Agent Development Kit (ADK) and are running into a persistent connectivity issue when trying to use the VertexAiSearchTool with our Vertex AI Data Store. We've followed the documentation and have tried numerous configurations, but we seem to be stuck and would greatly appreciate any insight from the community or the ADK team.

Our Goal:
Our objective is to build a simple agent that can query our private Confluence knowledge base. We have successfully set up a Data Store in the eu multi-region using the built-in Confluence connector in the Agent Builder UI.

The Problem:
No matter how we configure the resource path, any attempt by the agent to use the VertexAiSearchTool results in a 404 NOT_FOUND error. We've confirmed our user account has Owner permissions and that the Vertex AI and Vertex AI Agent Builder APIs are enabled.

The error log always shows that the API correctly resolves our Project ID to its corresponding Project Number, but the data store resource itself cannot be found:

google.genai.errors.ClientError: 404 NOT_FOUND. {'error': {'code': 404, 'message': 'DataStore projects/PROJECT_NUMBER/locations/.../dataStores/DATASTORE_ID not found.', 'status': 'NOT_FOUND'}}

---

What We Have Tried: Systematically Testing All Resource Paths

Our Data Store ID is confluence-vertexai_... (sanitized). We have systematically tested the following three formats for the DATA_STORE_ID in our .env file, none of which have worked.

1. Path based on the Console URL (assuming a custom collection):

• projects/our-project-id/locations/eu/collections/confluence-vertexai_.../dataStores/confluence-vertexai_...
• Result: 404 NOT_FOUND

2. Path using default_collection and the native eu location:

• projects/our-project-id/locations/eu/collections/default_collection/dataStores/confluence-vertexai_...
• Result: 404 NOT_FOUND

3. Path using default_collection and the global location:

• projects/our-project-id/locations/global/collections/default_collection/dataStores/confluence-vertexai_...
• Result: 404 NOT_FOUND

---

Our Current Configuration

Our setup is straightforward and aligns with the ADK documentation.

• agent.py:

  import os
  from google.adk.agents import Agent
  from google.adk.tools import VertexAiSearchTool
  
  DATA_STORE_ID = os.environ.get("DATA_STORE_ID")
  
  vertex_search_tool = VertexAiSearchTool(data_store_id=DATA_STORE_ID)
  
  root_agent = Agent(
      name="confluence_knowledge_agent",
      model="gemini-1.5-flash",
      instruction="You must use the VertexAiSearchTool to answer questions about internal policies.",
      tools=[vertex_search_tool],
  )

• .env File:

  GOOGLE_GENAI_USE_VERTEXAI=TRUE
  GOOGLE_CLOUD_PROJECT="our-project-id"
  GOOGLE_CLOUD_LOCATION="europe-west3" # For the Gemini model endpoint
  
  # This value was systematically replaced with each of the three paths listed above.
  DATA_STORE_ID="..." 

Our Questions for the Community

Since we've exhausted the obvious pathing solutions, we're wondering if we're missing a more fundamental concept.

1. Is there a known issue or a specific IAM permission required for eu multi-region data stores? We're curious if the service might return a 404 for a permissions issue instead of a 403.

2. Does the VertexAiSearchTool support "standalone" Data Stores? This is our main question. We created our Data Store directly in the Agent Builder UI. It is not part of a formally created "App" that has an associated "Serving Config". Does the tool require the data store to be explicitly attached to a serving configuration to be addressable via the API?

We're at a bit of a dead end and would be very grateful for any advice or clarification. Thank you!

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: The VertexAiSearchTool is designed to work with standalone Data Stores, and you do not need an "App" or "Serving Config". The 404 NOT_FOUND error is almost certainly due to an incorrect resource path for your Data Store. The most likely issue is the collection part of the path; it may not be default_collection.

---

Hi @this-dave,

Thank you for the detailed breakdown of the issue and the steps you've already taken. I can provide some clarification on your questions based on the ADK's design and documentation.

1. Does the VertexAiSearchTool support "standalone" Data Stores?

Yes, absolutely. The VertexAiSearchTool is designed to directly query a Vertex AI Data Store using its unique Data Store ID. [1, 2] There is no requirement for the Data Store to be part of a larger "App" or to have a "Serving Config". The tool is intended to provide a direct grounding source for your agent from your specified data store.

2. Is there an issue with eu multi-region or IAM permissions?

It's unlikely that the eu multi-region is the root cause of the problem. The resource path format is designed to accommodate different locations. While a 403 Forbidden error would more clearly indicate a permissions issue, a 404 Not Found error almost always means that the resource path you are using does not exist.

Recommendation: How to Find the Correct Resource Path

The 404 error strongly suggests that one of the components in your DATA_STORE_ID path is incorrect. Based on our documentation and examples, the format is indeed:

projects/{project_id}/locations/{location}/collections/{collection_id}/dataStores/{data_store_id} [3]

The most common point of error is the collection_id. You've correctly tried default_collection, but for a Data Store created directly in the Agent Builder UI, especially in a multi-region setup, the collection ID might be different.

To find the exact, full resource path, I recommend using the gcloud command-line tool. You can list all your data stores, which will provide the full and correct resource name to use:

gcloud vertex ai data-stores list --region=eu

This command should output the full name of your data store, including the correct project, location, and collection ID. You can then copy this entire path into your .env file.

I hope this helps you resolve the connectivity issue!

---

References
[1] Vertex AI Search Grounding Documentation
[2] VertexAiSearchTool Source Code
[3] Vertex AI Search Tool Snippet