A minimalist engine to ingest, connect, and retrieve your organization's collective intelligence. Reimagining Vannevar Bush's 1945 vision with geometric precision.
The primitives for a persistent graph of cross-referenced Neurons — compiled by an LLM, curated by you, navigated by association.
Sources aren't chunked and indexed — they're read by an LLM that authors a persistent wiki on your behalf. Every new fact updates existing Neurons instead of spawning isolated fragments.
Every Neuron (wiki page) links bidirectionally to its source documents and to related Neurons. Follow a claim back to its origin, or forward to everything it touches — the associative trails Vannevar Bush imagined.
Built-in SQLite FTS5 with auto-sync triggers keeps sources and compiled Neurons searchable in lockstep. No vector index to rebuild, no embedding drift.
Storage, LLM, vision, and pipeline processors sit behind pluggable interfaces. Start local with SQLite and Claude Code; swap in R2, Anthropic API, or a CMS consumer (@webhouse/cms, Laravel) without touching the engine.
Connectors roadmap: Slack, Discord, Notion, GitHub, Linear. Today: Markdown and PDF upload, plus native MCP for Claude Code and Cursor.
The LLM proposes, you dispose. Twelve candidate kinds, an auto-approval threshold, and a replay-able event log keep every change reviewable and reversible.