pythagora-io/gpt-pilot

GPT Pilot was an AI-powered tool that aimed to act as a real software developer rather than just a code autocomplete assistant. Instead of suggesting the next line of code, it could take a plain-English description of an app you want to build and then write, test, and debug the entire thing, only pausing to ask clarifying questions or request human review when needed. The README notes this repository is no longer maintained and the project has moved to a commercial product at Pythagora.ai. The system works by orchestrating a team of specialized AI agents, each playing a role like those found on a real development team. A Specification Writer agent refines your requirements, an Architect agent selects technologies and ensures they are installed, a Tech Lead agent breaks work into tasks, a Developer agent plans each task in plain language, a Code Monkey agent actually writes the code, a Reviewer agent checks it, and a Debugger agent handles problems when they arise. Each agent calls a large language model (LLM) such as GPT-4 or Claude under the hood, and all progress is saved to a database so work can be paused and resumed across sessions. Someone would use this when they want to build a complete working web application with minimal manual coding, describing what they want rather than writing it themselves. It is aimed at developers who want to delegate the tedious parts of scaffolding and boilerplate, while still staying in control of the high-level design. The tech stack is Python 3.9 or higher, with SQLite as the default database and optional PostgreSQL support. It runs as a command-line tool or as a VS Code extension.