Anyone was able to make the LiteLLM+Gemini+Tools combination to work?

[dpolist]

I was able to make LiteLLM+OpenAI+Tools to work perfectly fine. For some reason, Gemini (Vertex or not) doesn't. It stops after detecting the tool call. A call like this

agent("Hello, how are you? How much is 2+2")

Renders the following output:

I'm doing well, thank you!


Tool #1: calculator
➜  strands-lab git:(main) ✗ 

[yonib05]

Hey @dpolist, apologies for the extremely stale reply here. This one is old enough that a lot has changed since you asked, mostly in your favor. The biggest update: Strands now ships a native Gemini provider, so you don't need to go through LiteLLM for Gemini at all. Install with pip install 'strands-agents[gemini]' and use it directly:

from strands.models.gemini import GeminiModel

model = GeminiModel(
    client_args={"api_key": "<KEY>"},
    model_id="gemini-2.5-flash",
)

Tool calling works through the standard agent loop with this provider. Docs are here: Google provider.

If you specifically need to stay on LiteLLM (for example behind a LiteLLM proxy), the LiteLLM provider has had several fixes to streamed tool call handling since mid-2025, so upgrade both strands-agents and litellm to current versions and retry. If you still see the agent stop right after the tool call on current versions, please open a bug report with the model ID and versions so we can dig in.