// the find
Nikita-Filonov/ai-review
🚀 AI-powered code review tool for GitHub, GitLab, Bitbucket Cloud, Bitbucket Server, Azure DevOps and Gitea — built with LLMs like OpenAI, Claude, Gemini, Ollama, Bedrock, OpenRouter and Azure OpenAI
A CI/CD-integrated AI code reviewer that posts inline and summary comments on pull requests across GitHub, GitLab, Bitbucket, Azure DevOps, and Gitea. Supports seven LLM backends and an agent mode where the model can shell out to explore the repo before writing the review. Aimed at teams that want automated first-pass review without hosting their own infrastructure.
The VCS abstraction is genuinely broad — six platforms with a consistent adapter pattern means adding a new one is mechanical rather than a rewrite. Agent mode using ReAct-style shell tool calls (ls, cat, rg, git) gives the model actual repo context instead of reviewing a diff in a vacuum, which addresses the biggest failure mode of single-shot LLM review. The reply modes that let the AI participate in existing threads are a real workflow improvement over tools that just dump a top-level comment and walk away. Cost tracking and a pricing.yaml file suggest the author thought about the operational side, not just the happy path.
The agent mode gives the model shell access inside your CI runner — there is no sandboxing described, so a sufficiently adversarial prompt in a PR diff could instruct the agent to run arbitrary commands. The review quality is entirely dependent on the prompt templates, which are plain markdown files with no versioning or evaluation harness; you have no way to tell if a prompt change made reviews better or worse. There is no rate limiting or batching strategy visible for large PRs — sending 200 file diffs to an LLM in one shot will either hit context limits or cost more than the PR is worth. The package name on PyPI is xai-review rather than ai-review, which is a silent footgun waiting to cause install confusion.