AGENT_ENGINE_FAILED_PRECONDITION on testing Vertex AI Agent in Gemini Enterprice

Google Cloud Build with AI Agents
1

Hi All

I have an agent in Agent Engine. I’m able to test it trough the Playground and it works fine.

Now I’m trying to register it in the Gemini Enterprise.

I tried trough the UI, trough the make register-gemini-enterprise target and via the VeerMuchandi/agent_registration_tool tool. I can see the agents are registered but when I test them trough the Gemini Enterprise I get the following error:

{
“error”: {
“code”: 400,
“message”: “Agent finished execution without providing any response.”,
“status”: “FAILED_PRECONDITION”,
“details”: [
{
“@typetypetypetypetypetypetypetype”: “type .googleapis.com/google.rpc.ErrorInfo”,
“reason”: “AGENT_ENGINE_FAILED_PRECONDITION”,
“domain”: “discoveryengine . googleapis . com”
},
{
“@type”: “type.googleapis. com / google.rpc.RequestInfo”,
“requestId”: “assist_token: (PII Removed by Staff)
}
]
}
}

My adk is version 1.18.0.

How do I debug this? I can’t see any related sessions in the Agent Engine. In the error logs all I see is ValueError: Session not found …

Thank you for your time having a look at this.

2

got it working. Here are the steps:

  1. adk create agent

  2. do a standard deploy, not the accelerated one. Agent Engine - Agent Development Kit

  3. register the agent trough the gemini enteprise ui. I waited for 3 weeks to get all the UI options rolled out.

  4. use latest python and adk

  5. use us-central1 region

3

I see same error, I am already using
google-cloud-aiplatform[agent_engines,adk]>=1.127.0 and google-adk>=1.18.0

Using us-central1 for agent engine, but Enterprise app is global

I am using Agent Engine deployment steps using agent_engine.create not express deployment.

In Log explorer this’s what I see

Traceback (most recent call last):\n File \“/usr/local/lib/python3.10/threading.py\”, line 1016, in _bootstrap_inner\n self.run()\n File \“/usr/local/lib/python3.10/threading.py\”, line 953, in run\n self._target(*self._args, **self._kwargs)\n File \“/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/vertexai/preview/reasoning_engines/templates/adk.py\”, line 1042, in _asyncio_thread_main\n asyncio.run(_invoke_agent_async())\n File \“/usr/local/lib/python3.10/asyncio/runners.py\”, line 44, in run\n return loop.run_until_complete(main)\n File \“/usr/local/lib/python3.10/asyncio/base_events.py\”, line 649, in run_until_complete\n return future.result()\n File \“/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/vertexai/preview/reasoning_engines/templates/adk.py\”, line 1015, in _invoke_agent_async\n for event in runner.run_async(\nTypeError: 'async_generator’ object is not iterable",

4

Same issue here unable to get any of the agents integrated. Latest one I tried was the ADK using the cookiecutter project as per doc and it fails as well
Traceback (most recent call last):
File “/code/app/api/factory/python_file_api_builder.py”, line 473, in _async_stream_json_generator
async for chunk in output:
File “/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/vertexai/agent_engines/templates/adk.py”, line 1117, in streaming_agent_run_with_events
async for event in self._tmpl_attrs.get(“in_memory_runner”).run_async(
File “/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/google/adk/runners.py”, line 443, in run_async
async for event in agen:
File “/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/google/adk/runners.py”, line 380, in _run_with_trace
raise ValueError(message)
ValueError: Session not found: temp_session_2b6205c698ad11dd

5

I also have the same exact issue, trying to connect to a remote MCP server via the gemini enterprise authentication, but the agent fails with the same error.

Anyone found a fix?

6

From GCP Support , I got this answer: - To fix the issue, our product specialists recommend locking the package versions by google-cloud-aiplatform[adk, agent_engines]<=1.126.1 and redeploy the agent then register to Gemini Enterprise to see if the code can work. You confirmed that a different version combination resolved the issue on your end. Initially, yesterday with said configuration it didnt work for me, may be my agent was not rebuilding fine (?) but later it worked with these versions

1 Like
7

Thanks for your reply. Unfortunately my issue still persists, probably due to some code I added to handle the Oauth process for an MCP tool I’ve been trying to set up - can’t quite figure this out

8

Hey mate, the way to fix it, not sure why it fixes it yet but try the following
Install the latest version of the adk

run adk create agent

then adk deploy

For some reason hooking up this agent works

1 Like
9

That actually worked… But how and why? :sweat_smile:

10

[Reasoning Engine] Session initialization failed when ADK agent invoked from Gemini Enterprise

Problem

My ADK agent deployed on Vertex AI Agent Engine works fine when tested through Playground and direct API calls (streamQuery), but fails with Session initialization failed when called from Gemini Enterprise.

This was working yesterday. No code or config changes were made on our side.

Setup

  • Project: (PII Removed by Staff)

  • Reasoning Engine: (PII Removed by Staff)

  • Discovery Engine app: (PII Removed by Staff)

  • Agent registration ID: (PII Removed by Staff) (state: ENABLED)

  • Tested with google-adk 1.23.0, 1.24.0, and 1.25.0 — same result

  • google-cloud-aiplatform 1.137.0

Error

From reasoning_engine_stderr logs:

root - ERROR - _async_stream_json_generator - Error during async stream generation: Session initialization failed.

Stack trace points to Google’s managed runtime, not our code:

File "vertexai/agent_engines/templates/adk.py", line 1212, in streaming_agent_run_with_events
    raise RuntimeError("Session initialization failed.")

The request hits Agent Engine with 200 OK, then fails during streaming when the runtime tries to initialize a session internally.

Gemini Enterprise returns an empty response with "state": "SUCCEEDED" but no actual answer content. Old conversations now show “This conversation is read-only as the agent used is no longer available.”

What works

  • Creating sessions via the sessions API works fine

  • Calling the agent directly via streamQuery works — agent executes, calls tools, returns results

  • Playground works with full execution logs