Brain2

A single-user work knowledge system that converts messy inputs - handwritten notes, pasted text, PDFs, voice memos and forwarded email - into structured, reviewable long-term work memory.

The pipeline is capture, parse, review, save, export: an input is uploaded or pasted, OpenAI extracts structured data in a single vision call that handles OCR and extraction together, the user edits every field on a review page, and only then is it written to Supabase and auto-exported as markdown.

Extracted content is decomposed into linked entities - notes, tasks, people, projects, domains, decisions, open questions and ideas - each with its own list and detail view, joined through cross-link tables.

Semantic search runs over entity embeddings, and an AI linking pass suggests connections between notes that were never explicitly related, turning a capture tool into something that surfaces its own connections.

Merge and dedup tooling for people, projects and domains keeps the graph clean as real data accumulates, alongside a maintenance dashboard and compiled wiki summaries per entity.

Work context was scattered across handwritten notebooks, chat transcripts and half-finished documents, none of which were searchable once the week had moved on.

Existing note apps store what you wrote; the gap was turning an unstructured capture into linked entities - who was involved, what was decided, what is still open - without retyping it all by hand.

The non-negotiable was that nothing becomes canonical until a human reviews and confirms it. AI drafts the structure, the user owns the record.

Wanted one place where a photo of a notebook page, a voice memo on the way home and a forwarded email all land in the same reviewable inbox and end up in the same graph.