Mesrai

Mesrai is a multi-agent AI code review tool for engineering teams, built around five specialized reviewers — Bug, Security, Performance, Architecture, and a custom Playbook agent — plus an AST-based context graph that reads your whole codebase's structure, not just the diff in front of it. In a 24-PR benchmark against CodeRabbit, Mesrai caught 75% of all issues and 86% of critical bugs, versus CodeRabbit's 63% and 57%. It runs BYOK (bring your own API key), so there's no per-seat rate-limiting or vendor markup — you pay for the model you use, plus $6/dev/month for Mesrai itself, roughly a fifth of what CodeRabbit ($24–30/dev) or Greptile ($30/dev) charge. It ships as a GitHub integration, a VS Code extension, a CLI, and 14 MCP plugins, with a marketplace of 800+ review rules. It also learns per-team: dismiss a finding type once, and Mesrai stops raising it in future reviews — no retraining, no config file. Key features / differentiators, if the form wants them separate: 5 specialized agents (not one generalist model) — Bug, Security, Performance, Architecture, Playbook AST context graph for whole-codebase awareness, not diff-only review 86% critical-bug catch rate vs. 57% for the category leader, on a 24-PR audit BYOK pricing — $6/dev/month, no markup on model costs Dismiss-and-learn: stops re-flagging finding types you've already rejected VS Code + CLI + 14 MCP plugins + 800+ rule marketplace