Add an option to limit number of tool calls

[Kipok]

Summary by CodeRabbit

• New Features
  • Added configurable maximum tool call limit for generation. Users can now set a maximum number of tool calls allowed; when reached, generation stops and returns "tool_call_limit_reached" status. Unlimited tool calls remain the default when not configured.

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (1)

nemo_skills/inference/model/tool_call.py
Tool call limit breaks state

When max_tool_calls is reached you break after appending generation["serialized_output"] (which includes the assistant tool-call message) but before appending the corresponding tool outputs. This leaves result_steps["conversation"] ending with an unresolved tool call, which can break downstream consumers that expect tool calls to be paired with tool responses (and the TODO notes this).

To keep the conversation consistent, avoid appending the tool-call message when you intend to stop, or alternatively execute only the remaining allowed tool calls (and include their outputs) instead of stopping mid-turn.

[coderabbitai]

📝 Walkthrough

Walkthrough

A new max_tool_calls configuration parameter is introduced to limit tool executions during generation. The parameter flows from GenerationTaskConfig through get_tool_calling_model to ToolCallingWrapper, where enforcement logic checks the limit and halts tool execution when reached.

Changes

Cohort / File(s)	Summary
Tool Call Limiting Feature nemo_skills/inference/generate.py, nemo_skills/inference/model/__init__.py, nemo_skills/inference/model/tool_call.py	Added max_tool_calls: int = -1 parameter to GenerationTaskConfig and threaded it through the model initialization pipeline. Implemented tool call counter and limit enforcement in ToolCallingWrapper.generate_async to track executed calls and halt execution with "tool_call_limit_reached" finish reason when limit is exceeded.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The pull request title accurately describes the main change: adding a configuration option to limit the number of tool calls during generation across three files in the inference/model subsystem.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch igitman/limit-tool-calls

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}