LiveReview

About LiveReview

LiveReview is a code review automation tool that risk-scores every pull request and auto-approves the lowest-risk changes. It combines an LLM, incident-trained machine learning, and code-graph analysis to route risky PRs to the engineers who should review them. The tool is free and can be self-hosted on your own infrastructure.

Review

LiveReview takes a production-outcome approach to code review-it measures success by incident reduction and rollback frequency, not comment volume. The tool draws from Meta's internal RADAR system, which the creators cite as delivering a 50x drop in production incidents and 3x fewer rollbacks. It's available now as a free, self-hostable Docker image or via a cloud version for evaluation.

Key Features

• Risk scoring and auto-approval: Each PR receives a risk score based on roughly 20 signals, including AST-based blast radius analysis, production feedback, code complexity, contributor trust levels, and LLM semantic feedback. The lowest-risk 30-50% of PRs are auto-approved; the rest always go to a human reviewer.
• Risk-first review queue: High-risk PRs are prioritized at the top of the review queue, with comments sorted by blast radius. This directs senior attention where the tool's scoring indicates it's most needed.
• Cognitive Coverage: Developers can walk through a PR as slides, read a summary, or take a quiz on the changes before merging. The feature aims to keep the team familiar with AI-generated code without requiring line-by-line manual review.
• Livi Assistant: A natural-language interface lets you query engineering data-adoption rates, ROI metrics, and team-level trends-across more than 50 graphs. It works on the web and through Slack, Teams, or Discord.
• Multi-platform and multi-model support: Integrates with GitHub, GitLab, Bitbucket, Gitea, and Azure DevOps. Users can bring their own API keys for Claude, OpenAI, or any of 15+ AI providers, or use the LiveReview-provided model.

Pricing and Value

LiveReview is free and self-hostable. Organizations run it on their own infrastructure using a Docker image available on GitHub. A cloud version exists for quick evaluation. Users can supply their own LLM API keys-Claude, OpenAI, or other supported providers-or rely on the LiveReview model. There is no listed paid tier or usage-based fee at this time.

Pros

• Auto-approval is gated on a risk score, not a blanket rule-low-risk PRs skip review while everything else still requires a human.
• Learns from production incidents and rollbacks over time, refining which code areas it flags as high-risk.
• Self-hosting and bring-your-own-model support mean code never has to leave your infrastructure or pass through a third-party reviewer.
• Cognitive Coverage gives teams structured ways to understand AI-generated changes without reading every diff.
• Integrates with five major source control platforms and 15+ AI providers, reducing vendor lock-in.

Cons

• The ML component that learns from production incidents needs historical data to become effective-teams without incident or rollback history won't benefit from that signal initially.
• Auto-approving even 30% of PRs requires organizational trust in the risk model; teams with strict compliance or audit requirements may find this unacceptable regardless of accuracy.
• LiveReview is not well suited for very small teams or solo developers who want a simple linting or style-checking tool-its design assumes a multi-reviewer queue and production feedback loop that smaller setups lack.

LiveReview fits teams that already have a review bottleneck and enough production history for the ML component to learn meaningful patterns. The self-hosted, bring-your-own-key model may appeal to organizations with strict data residency or security requirements. For teams still building their first review process or those operating under rigid manual-approval mandates, the auto-approval mechanism could introduce more friction than it removes.