What sets Peek apart from most other solutions is that all queries and results are laid out on an infinite canvas (like Miro or Figma). You can click on a primary key and see all references to that specific row, or click on a foreign key and it'll fetch and display those rows in new result nodes, kind of building a direct acyclic graph as you go. It'll also give you chart nodes for numerical data in results and you can draw on the canvas like a whiteboard or drop text nodes. You could also run two or three queries and check the results side by side.

There are a bunch more features such as fully P2P multiplayer (based on [iroh.computer](https://iroh.computer)) that works just like it does in figma/miro with live cursors etc. Guests never see the connection details, though, but rather they send commands to the host and the data is synced back to the guest when the host canvas updates. There is no centralized Peek service either, even for address resolution so you don't have to worry about leaking connection information, table data or anything else when using the multiplayer features.

There is a built-in AI agent that works both with local inference with your self-hosted OpenAI compatible server (ollama, llama.cpp etc) so nothing leaves the machine, or via MCP so Claude Code can create nodes, move the camera, analyze results etc. All AI features are fully optional of course and the MCP is disabled by default.

Have a Peek!