Introduction

peek turns a GitHub repository into a hierarchical atlas your AI can actually read — and gives you a small team of agents that knows how to use it.

Point a coding agent at a large codebase cold and it tends to wander: list the files, open a few at random, grep on a hunch, burn tens of thousands of tokens, and still answer from the wrong corner of the repo. peek fixes the part that's actually broken — the map. We summarize every file into its folder, every folder into a root map, and keep that atlas fresh as you push. A question then starts from a bird's-eye view and drills down to the exact lines, with citations.

Two ways to use it

• Chat on the web. Pick one or more repos and ask questions in plain English. You get a synthesized, source-cited answer streamed back live.
• From your own agent. peek runs a remote MCP server, so MCP-aware clients like Claude Code, Cursor, or Copilot can call it as a native tool — no copy-pasting code into a chat window.

Either path runs the same atlas and the same research team. Index once, ask anywhere.

Keep reading

• How peek works — the atlas and the research team, end to end.
• Working with repositories — adding repos, what to exclude, and how reindexing stays cheap.
• Pricing, credits & usage — pay only for the tokens your questions actually use.