cristicretu/AI Code Reviewer

Automated PR code review using a fine-tuned LLM, RAG over your codebase, and an agentic tool loop.

Trust Signals

name	description	required	default
model-api-base	OpenAI-compatible endpoint URL hosting the review model (e.g. a vLLM/TGI server serving cretu-luca/code-reviewer-grpo).	yes	—
model-id	Model identifier the endpoint expects (e.g. cretu-luca/code-reviewer-grpo).	yes	—
model-api-key	Bearer token for the model endpoint. Use any non-empty string if your endpoint does not require auth.	no	dummy
github-token	Token used to fetch the PR and submit the review. Needs pull-requests:write.	no	${{ github.token }}
max-agent-steps	Hard cap on agent tool-loop iterations.	no	20
comment-budget	Maximum number of inline review comments to post per PR.	no	10
python-version	Python version used inside the action.	no	3.11

no outputs