shlokkhemani/rabbithole

What it solves

Rabbithole provides a non-linear, infinite canvas for learning and exploration. It prevents the loss of context when diving deep into specific topics by allowing users to ask questions at any point in a document and branching off into new, linked documents based on the answers.

How it works

The system operates through a canvas that can be hosted in two ways: as a static web app (connecting to various model endpoints like OpenRouter or OpenAI-compatible APIs) or via an MCP server. The MCP server allows AI agents like Claude Code or Codex to interact with the canvas while keeping storage and transport local to the user's machine. Documents are stored locally in the browser or in a specific local directory, avoiding the need for a hosted document store or account.

Who it’s for

It is designed for learners and researchers who want to visualize their learning paths and maintain a structured yet flexible way to explore complex subjects through AI-driven questioning.

Highlights

• Infinite canvas interface for branching learning paths.
• Support for multiple model endpoints including OpenRouter and local OpenAI-compatible APIs.
• MCP server integration for use with AI agents like Claude Code and Codex.
• Privacy-focused design with no telemetry, no accounts, and local-only document storage.