DeepDive

DeepDive is a place to capture ideas, projects and tasks — and, above all, the connections between them. Instead of folders and lists, every item is a node on a 3D globe you rotate, zoom into and explore. Zoom out far enough and the globes become a navigable universe, each one a context of its own.

What the project aims to answer

Isolated notes age badly. What usually gets lost isn't the content but the relationship: why an idea existed, what it depended on, what it unlocked. DeepDive assumes that spatial structure and explicit connections preserve that context better than file hierarchies.

How it works

A node can be an idea or a task with a due date, and carries a block-editor description, attachments, category, tags and priority. Search spans title, tags, category and description, with ranking and a ⌘K shortcut. Pinned nodes stay in the foreground — the globe's camera drifts back to them on its own. A tree panel lists everything by category, for when the 3D view isn't the fastest route. Extra globes can be shared with another account, turning a personal context into a collaborative workspace.

AI access

DeepDive exposes its own graph through a remote MCP server, so an assistant like Claude can read, create and connect nodes directly. The graph stops being just a place where you write things down and becomes working memory shared with the AI.

Architecture

A pnpm + Turborepo monorepo: Next.js front-end on Vercel, serverless AWS back-end (Cognito, DynamoDB, S3, Lambda, API Gateway) provisioned with CDK, and a Tauri v2 Windows desktop app running the same front-end as a borderless, always-signed-in window.