jellydn/ai-flow

AI Flow turns GitHub URLs into structured AI workflows without you having to write a prompt yourself. You paste a link to a public pull request, issue, or repository, pick one of several prebuilt launchers, and a queue worker gathers the relevant context, calls an AI provider, checks the response against a fixed format, and hands back a shareable report. The built-in launchers cover common engineering tasks. Review PR looks at a public pull request and points out risks, findings, and things worth double checking. Plan Issue turns a GitHub issue into a step-by-step implementation plan. Explain Repo summarizes a public codebase, covering its architecture, structure, and key patterns. Laravel Doctor inspects a Laravel project and suggests improvements around conventions, security, and performance. Signed-in users can also build their own custom launchers with their own names, prompts, and output formats. Every run produces a structured, schema-checked report rather than a block of free-form chat text, so each finding comes with a severity label, file references, and concrete fix suggestions, and results can be shared by URL. Progress streams to the browser in real time as the run executes. The backend is built with Laravel and PHP, the frontend with React and TypeScript, and runs are processed through a database-backed job queue. It can call OpenAI, OpenRouter, Anthropic, or Gemini as the AI provider, and stores data in SQLite for development or PostgreSQL and MySQL in production. Guests can use it right away through OpenRouter's free model router with no setup, while signed-in users can supply their own provider and API key. Provider keys are never stored on runs or returned by the API, and saved credentials are encrypted at rest. Setting it up involves cloning the repo, installing PHP and Node dependencies, running database migrations, and adding at least an OpenRouter API key before starting the app locally. The project is released under the MIT License.