llm-advisor

Use the llm CLI to send your problem description, stack trace, and troubleshooting steps to a reasoning model like o4-mini or Gemini 2.5 Pro. Pipe in relevant code or logs so the model has full context for its analysis.

Pipe the output of git diff into the llm CLI with a review prompt, for example targeting gpt-4.1 or claude-sonnet-4-5. Ask for bugs, security vulnerabilities, performance issues, and missing edge cases with specific line references.

For complex debugging use o4-mini with high reasoning effort or Gemini 2.5 Pro with a large thinking budget. For quick questions use gpt-4o-mini or gemini-2.0-flash, and for deep architecture reasoning use o3 or claude-opus-4-5.

Yes, install the llm-gemini and llm-anthropic plugins, then set API keys with llm keys set gemini and llm keys set anthropic. Run llm models to verify which model IDs are available after installation.

Model names change frequently as providers release new versions. Run llm models to list currently installed model IDs, and upgrade the llm package and its plugins with pip install --upgrade to get the latest model names.

Avoid it for simple obvious changes like typos, well-documented operations with clear solutions, and low-risk changes where speed matters. Always verify suggestions against your codebase and official documentation since models can be wrong about recent APIs.