grepai

Why we included this project

Your coding agent eats context tokens on every prompt, and grepai attacks that directly. It indexes your repository with vector embeddings, so you can ask for code by intent: a query like 'authentication flow' surfaces the session handler even when the file name gives nothing away. That reduces the input an assistant needs to reason about your codebase, and the savings show up immediately. It also builds call graphs, so you can see every caller before refactoring a function, and a file watcher keeps the index fresh without manual rebuilds. Everything runs locally by default with Ollama or LM Studio (OpenAI is an option), and it ships as an MCP server, so Claude Code, Cursor, or Windsurf can call it directly as a tool instead of you pasting snippets.