build_graph 500: serialize_agent fails to JSON-serialize LiteLlm model (LiteLLMClient)

[rcchirwa]

Describe the Bug:
The dev UI's agent-graph panel calls GET /dev/apps/{app}/build_graph, which returns 500 for any agent whose model is a LiteLlm instance. serialize_agent calls model_dump() on the model field, which embeds the live LiteLLMClient (LiteLlm.llm_client) — a non-JSON-serializable object — so FastAPI's response serialization raises PydanticSerializationError. Affects every LiteLLM-backed agent.

Steps to Reproduce:

1. Install google-adk==2.1.0.
2. Minimal, framework-only reproduction:

   import json
   from google.adk.agents import LlmAgent
   from google.adk.models.lite_llm import LiteLlm
   from google.adk.cli.utils.graph_serialization import serialize_agent
   
   agent = LlmAgent(name="repro", model=LiteLlm(model="ollama_chat/llama3"))
   json.dumps(serialize_agent(agent))
   # TypeError: Object of type LiteLLMClient is not JSON serializable

3. End to end: run adk web with a LiteLlm agent, open the agent in the UI; the graph panel fires GET /dev/apps/{app}/build_graph -> 500.

Expected Behavior:
build_graph returns a JSON graph of the agent; the model field is summarized (e.g. model name/string) rather than dumping the runtime client.

Observed Behavior:
500 Internal Server Error. Stack trace tail:

File ".../fastapi/_compat/v2.py", line 231, in serialize_json
    return self._type_adapter.dump_json(...)
File ".../pydantic/type_adapter.py", line 677, in dump_json
    return self.serializer.to_json(...)
pydantic_core._pydantic_core.PydanticSerializationError: Unable to serialize unknown type: <class 'google.adk.models.lite_llm.LiteLLMClient'>

Root cause: cli/utils/graph_serialization.py serialize_agent -> the generic else branch does value.model_dump() on the model field; LiteLlm.model_dump() includes llm_client: LiteLLMClient. The /dev/apps/{app}/build_graph route returns this raw (-> Any, no response model), so FastAPI then can't JSON-encode it. (Note: the route is already marked # TODO: remove this endpoint once build_graph_image is completed; build_graph_image returns 200.)

Environment Details:

• ADK Library Version (pip show google-adk): 2.1.0
• Desktop OS: macOS 26.5 (arm64)
• Python Version (python -V): 3.11.15

Model Information:

• Are you using LiteLLM: Yes (litellm 1.86.2)
• Which model is being used: ollama_chat/* via LiteLlm (e.g. ollama_chat/gemma4)

[jordanchendev]

Hi maintainers, I'd like to work on this issue if that's okay.

I reproduced the failure locally with a regression test for LlmAgent(model=LiteLlm(...)): serialize_agent() currently includes LiteLlm's runtime llm_client object in the graph payload, which makes the result fail JSON serialization.

The local fix I'm testing is small: exclude the runtime llm_client field when graph serialization dumps model config, while keeping the configured LiteLlm model name in the graph response. I also added a unit test under tests/unittests/cli/utils/test_graph_serialization.py to assert the serialized graph is JSON-safe.

If this direction looks acceptable, I can open a PR linked to this issue.

[boyangsvl]

/adk-issue-analyzze

[llalitkumarrr]

Hello @rcchirwa ,

We were able to reproduce the issue and observed the same behavior. To help resolve this, could you please update line 2184 in src/google/adk/models/lite_llm.py?

Please change:
llm_client: LiteLLMClient = Field(default_factory=LiteLLMClient)

to:
llm_client: LiteLLMClient = Field(default_factory=LiteLLMClient, exclude=True)
Could you please test this change and let us know if it resolves the issue for you?