feat: Gemini toolcall

[JINO-ROHIT]

Gemini support structured output with tool call.

Links #2364

[Wendong-Fan]

thanks @JINO-ROHIT , With more complex tasks I’m seeing errors when we force the “tool‑call first, then structured output” sequence. Sometimes ChatAgent().step makes multiple LLM calls, which breaks that assumption. A cleaner fix might be to move the handling into ChatAgent()._format_response_if_needed instead?

[Wendong-Fan]

use await for async function

[JINO-ROHIT]

hmm great suggestion, will do

[JINO-ROHIT]

can you also post a sample complex task where it fails, so i can test after my changes, thanks!

[Wendong-Fan]

can you also post a sample complex task where it fails, so i can test after my changes, thanks!

https://github.com/camel-ai/camel/blob/master/examples/structured_response/json_format_reponse_with_tools.py

[JINO-ROHIT]

honestly, ive tried a lot to make this work, but in general gemini tool calling seems awful, i cant really get it to work.
(in general what i read from the internet is also along the same lines)

any ideas on what we do next?

[Wendong-Fan]

honestly, ive tried a lot to make this work, but in general gemini tool calling seems awful, i cant really get it to work. (in general what i read from the internet is also along the same lines)

any ideas on what we do next?

No worries at all, Jino. I appreciate the effort you've put into the Gemini tool calling. I understand that it's tricky, and it's more of a 'nice-to-have' right now. Let's park this for the time being, and I'll take another look once I've cleared some higher-priority items from my plate