What is codepilot-ai?

CodePilot AI is a self-hosted tool that automatically reviews code changes and gives feedback, similar to what a senior developer might leave in a pull request. What sets it apart from a typical AI reviewer is that it keeps a persistent memory of your team's coding conventions, which frameworks you use, and which past suggestions your team accepted or rejected, so its feedback is meant to improve the more it is used on a given codebase rather than starting fresh each time. To keep costs down, the system routes each code review through a smaller, cheaper AI model first, and only sends the code to a larger, more capable model when the review is complex or touches something security sensitive. The README states this approach can cut costs significantly compared to always using the most powerful model. Alongside AI review, the tool also runs traditional static code analysis using Tree-sitter and Semgrep to catch structural and security issues directly, not just through AI judgment. Every review produces a score broken into categories such as correctness, security, performance, style, testing, and documentation, and the interface includes a dashboard showing how much money the smart routing has saved over time. There is also a feedback loop where a developer can accept, reject, or ignore each suggestion, and those choices feed back into the memory system to refine future reviews. The project is built with a Next.js frontend, a FastAPI backend written in Python, a PostgreSQL database, and Redis for caching, all wired together and run through Docker Compose. To try it, you need Docker installed and a free API key from Groq, the AI provider it uses for inference. This is a developer tool aimed at engineering teams who want automated code review integrated into their workflow rather than an individual hobby project.