An experiment in giving a company one shared institutional brain through memory, retrieval, and graph-based context.
What it is
Crystal is an experiment in what it would look like for a company to have one shared institutional brain. It ingests communications from Slack and Google Drive, extracts decisions and entities via LLM, and makes that knowledge queryable through a natural language interface and an interactive knowledge graph.
The idea is that organizations already think in fragments: Slack threads, Drive docs, meeting notes, decisions, handoffs, and half-remembered context. Crystal treats those fragments as pieces of one living memory system, exploring whether a company can develop a persistent, searchable layer of shared understanding.
Stack
Dual-database architecture: Neo4j for the knowledge graph (relationships between people, decisions, entities) and pgvector for semantic search. Async ingestion pipeline, real-time Slack bot integration, Next.js frontend.
Status
Currently in active development with Alexi as an exploration of institutional memory, organizational context, and AI systems that help teams remember what they collectively know.