RabbitHoles AI

Provides a spatial, non-linear chat interface where conversations expand across an infinite 2D canvas rather than a linear message thread. Users can position conversation nodes, branches, and AI responses spatially, creating visual mind-map-like structures. The canvas supports pan, zoom, and spatial organization of dialogue history, enabling users to explore multiple conversation threads simultaneously and navigate between them by spatial position rather than chronological order.

Maintains conversation context and AI state across spatially-separated conversation branches on the canvas. When users branch a conversation by asking a new question in a different canvas location, the system preserves the prior conversation history and system context, allowing the AI to reference earlier discussion points while exploring new tangents. This requires maintaining a graph-based conversation state rather than linear message history.

Enables extended multi-turn conversations where each AI response and user follow-up can be positioned independently on the canvas. The system manages conversation flow across multiple turns while allowing users to interleave responses, ask questions about specific prior responses, or create new branches at any point in the dialogue. This requires stateful session management that tracks which response each follow-up question references.

Enables users to export or share conversation canvases in a format that preserves spatial layout, conversation structure, and context relationships. The system likely serializes the canvas state (node positions, connections, conversation content) into a shareable format that can be viewed, imported, or collaborated on. This requires a structured data format that captures both the conversation content and spatial metadata.

Provides AI-assisted search and navigation across the conversation canvas, allowing users to find relevant prior discussion points, jump to related topics, or get AI-generated summaries of specific canvas regions. The system likely uses semantic search or embeddings to match user queries against conversation content and spatial clusters, enabling intelligent navigation of large conversation trees without manual scrolling.

Streams AI responses token-by-token directly to canvas nodes as they are generated, providing real-time feedback without waiting for complete response generation. The system likely uses WebSocket connections or Server-Sent Events to push streaming tokens to the frontend, rendering them incrementally in the positioned canvas node. This enables users to see AI thinking in progress and interact with partial responses.

Provides UI tools for organizing conversation nodes spatially on the canvas, including pan, zoom, node repositioning, grouping, and potentially auto-layout algorithms. Users can manually arrange nodes to create visual clusters representing related topics, or use automatic layout suggestions to organize large conversation trees. The system likely supports drag-and-drop positioning, viewport management, and spatial queries for finding nearby nodes.