LiteLlm root agent stuck on greetings after a couple of dialog

[cl3m3nt]

** Please make sure you read the contribution guide and file the issues in the right place. **
Contribution guide.

Describe the bug
While using LiteLLM with gpt-oss-20b or gemma-3n as root agent, the adk web is stuck after a couple of dialogs.

To Reproduce
Please share a minimal code and data to reproduce your problem.
Steps to reproduce the behavior:
root_agent = Agent(
name="my_agent",
model=LiteLlm(
model="openai/gpt-oss",
api_base="http://localhost:1234/v1",
api_key="lm-studio"),
description=("..."),
instruction=("...")
)

• using gps-oss-20b in LM studio
• using ag-ui-adk 0.3.2, google-adk 1.17.0

Expected behavior
The root agent is gpt-oss-20b (or gemma-3n). After a couple of successful dialog, the agent is either looping on greetings, or always greetings. It is no longer able to answer something different than greeting nor call tools.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: macOS Apple Silicon M4
• Python version(python -V): Python 3.12.11

• ADK version(pip show google-adk): ag-ui-adk 0.3.2, google-adk 1.17.0

Model Information:

• Are you using LiteLLM: Yes
• Which model is being used: gpt-oss-20 or gemma-3n

Additional context
When using same models, gemma-3n or gpt-oss-20 through FastAI server + Gradio, there stuck behavior does not occur so far, while still serving models through LM studio and adk. To be noticed in those cases I am using session + runner, while not using session + runner with adk-web interface.

[cl3m3nt]

Additionnal info from LM studio logs error, as well as scenario where no answers at all providing within UI

[surajksharma07]

@cl3m3nt This greeting loop issue in v1.17.0 is caused by content filtering bugs in flows/llm_flows/contents.py where empty/thought-only events weren't properly filtered, confusing the agent's context.

The issue is fixed in v1.21.0+ through commits f35d129 (checks all content parts, not just first) and 07bb164 (excludes thought parts from agent output).

Upgrade to the latest ADK version to resolve: pip install --upgrade google-adk

If you can't upgrade immediately, use Runner with InMemorySessionService directly instead of adk web (as you mentioned works in your FastAPI+Gradio setup) to avoid the web UI's event handling issues.

Thanks!