oracle

Why we included this project

Oracle is a command-line tool that bundles a prompt with the files you select and sends that context to a frontier model, so the review is grounded in your actual code rather than a summary you wrote by hand. It talks to providers through their APIs, or it can drive a signed-in ChatGPT or Gemini browser session if you would rather skip setting up API keys. Before anything reaches a model, a render mode prints the exact prompt and the numbered file contents locally, which is handy for checking what you would actually send. Responses are stored as sessions, so long runs finish in the background and can be replayed or continued with follow-ups, and a multi-model mode logs per-model usage and cost together with the output in a single run. Teams using coding agents like Claude Code, Codex, or Cursor can wire Oracle in through its MCP server and use it as a second opinion inside their existing workflow.