LiteLlm drops Gemini grounding_metadata: _model_response_to_generate_content_response ignores response.vertex_ai_grounding_metadata

[adityapandey216]

🔴 Required Information

Please ensure all items in this section are completed to allow for efficient
triaging. Requests without complete information may be rejected / deprioritized.
If an item is not applicable to you - please mark it as N/A

Describe the Bug:

When using LiteLlm to call Gemini models with grounding enabled (google_search tool, Vertex AI Search, etc.), the resulting LlmResponse.grounding_metadata is
always None. Downstream consumers (event.grounding_metadata, after_model_callback, after_agent_callback) therefore have no access to grounding
chunks/supports, breaking citation pipelines that rely on byte-index supports.

The native Gemini() model class propagates grounding_metadata correctly. Only the LiteLlm path is affected.

Steps to Reproduce:

  import asyncio
  from google.adk.agents import LlmAgent
  from google.adk.models.lite_llm import LiteLlm
  from google.adk.runners import InMemoryRunner
  from google.adk.tools import google_search
  from google.genai import types

  agent = LlmAgent(
      name="grounded_agent",
      model=LiteLlm(model="gemini/gemini-2.5-flash"),
      instruction="Answer using web search.",
      tools=[google_search],
  )

  async def main():
      runner = InMemoryRunner(agent=agent, app_name="repro")
      session = await runner.session_service.create_session(
          app_name="repro", user_id="u"
      )
      async for event in runner.run_async(
          user_id="u",
          session_id=session.id,
          new_message=types.Content(
              role="user",
              parts=[types.Part(text="When was Google Inc. founded?")],
          ),
      ):
          if event.grounding_metadata:
              print("HAS GROUNDING:", event.grounding_metadata.grounding_chunks)
          else:
              print("event.grounding_metadata is None")

  asyncio.run(main())

Expected: At least one event prints HAS GROUNDING: [...] with non-empty grounding_chunks and grounding_supports (matching the behavior of the same code
with model=Gemini(model="gemini-2.5-flash")).

Actual: Every event prints event.grounding_metadata is None. The Gemini API returned grounding (verifiable via litellm._turn_on_debug()), LiteLLM stored it
on the ModelResponse, but ADK silently discards it.

Environment Details:

• google-adk: 1.25.1 (also reproduced against main — the conversion functions in lite_llm.py on main still contain zero references to grounding_metadata)
• litellm: any recent version
• Model: gemini/gemini-2.5-flash (also gemini/gemini-2.0-flash, vertex_ai/gemini-2.5-pro)
• Python: 3.12

Model Information:

• Are you using LiteLLM: Yes
• Which model is being used: gemini-3.1-flash-preview

---

🟡 Optional Information

Providing this information greatly speeds up the resolution process.

Regression:
Did this work in a previous version of ADK? If so, which one?

Logs:
Please attach relevant logs. Wrap them in code blocks (```) or attach a
text file.

// Paste logs here

Screenshots / Video:
If applicable, add screenshots or screen recordings to help explain
your problem.

Additional Context:
Add any other context about the problem here.

Minimal Reproduction Code:
Please provide a code snippet or a link to a Gist/repo that isolates the issue.

// Code snippet here

How often has this issue occurred?:

• Always (100%)
• Often (50%+)
• Intermittently (<50%)
• Once / Rare

[1wos]

Hi, would like to pick this up — taking a look now.

[llalitkumarrr]

Hello @adityapandey216 ,

We highly recommend using the Gemini model class instead of LiteLlm for all Gemini models. This class offers better integration, ensuring you can fully use all of Gemini's features. We also recommend upgrading to the latest ADK version (google-adk==1.33.0)

[adityapandey216]

Hey @llalitkumarrr,

Thanks for responding. We were previously accessing the Gemini model directly using the Gemini model class. However, due to strict organizational policies, all LLM models must now be accessed exclusively through LiteLLM. As a result, we are migrating our Gemini model access to go through LiteLLM as well.

[llalitkumarrr]

Hello @adityapandey216 ,

Since google_search tool is specific to Gemini, using it with LiteLLM may cause issues. This is why we highly recommend using the Gemini model class, which allows you to fully utilize all Gemini specific tools without any problems.