fix: remove global add_function_to_prompt — breaks native tool calling (Groq, OpenAI)

[vitas]

Problem

_ensure_litellm_imported() sets litellm.add_function_to_prompt = True globally at import time (line 188). This forces ALL models through LiteLLM's text-based tool calling path — tool definitions are injected into the system prompt as text instead of being passed as the tools parameter.

Models that support native function calling (Groq, OpenAI, Anthropic) then output XML-style function tags instead of proper tool_calls JSON:

<function=run_command {"command": "kubectl get pods -n demo"} </function>

Groq rejects with:

{"error":{"message":"Failed to call a function. See 'failed_generation'","code":"tool_use_failed"}}

Proof

Direct litellm.completion() inside the same environment without the global flag returns proper tool_calls JSON with finish_reason: "tool_calls":

# Works (direct LiteLLM, no global flag):
litellm.completion(model="groq/llama-3.3-70b-versatile", tools=[...])
# → finish_reason: "tool_calls", proper JSON ✓

# Fails (via ADK, which sets the global flag):
Agent(model=LiteLlm(model="groq/llama-3.3-70b-versatile"), tools=[...])
# → XML tags, Groq 400 ✗

Fix

Remove the global litellm.add_function_to_prompt = True. Models that need text-based tool calling (e.g., some Ollama models without native support) can opt in per-instance:

LiteLlm(model="ollama/qwen2", add_function_to_prompt=True)

This kwarg flows through _additional_args into acompletion(), so per-model opt-in already works.

Impact

• Fixes: All models with native tool calling (Groq, OpenAI, Anthropic) through LiteLlm
• No regression for models without native tool calling: They can pass add_function_to_prompt=True explicitly
• One line removed, comment added explaining why

References

• [BUG] Groq API fails to call tools for ToolCallingAgents huggingface/smolagents#1119 — same XML tag issue across frameworks
• Groq not working well with the Litellm definition of tools. BerriAI/litellm#11001 — Groq tool calling format issues
• Introduced in f33e090 for Support tool-calling for LLMs with coding skills #1273 (support models without native tool calling)

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[adk-bot]

Response from ADK Triaging Agent

Hello @vitas, thank you for creating this PR!

Before we can merge this, could you please:

1. Sign the Contributor License Agreement (CLA). You can find more information at https://cla.developers.google.com/.
2. Associate a GitHub issue with this PR. If there is no existing issue, could you please create one?
3. Add a testing plan section to your PR description to describe how you tested these changes.

This information will help reviewers to review your PR more efficiently. Thanks!

[rohityan]

Hi @vitas , Thank you for your contribution! We appreciate you taking the time to submit this pull request. Your PR has been received by the team and is currently under review. We will provide feedback as soon as we have an update to share.

[rohityan]

Hi @GWeale , can you please review this.