By the Museum of Crypto Art
Cortex is the memory layer for AI agents.
Agents tend to forget. Cortex is how your organization remembers — a managed platform that turns documents into a living knowledge graph agents can read, write, and reason over. Model- and framework-agnostic, born from the museum preserving its own institutional knowledge, and the technology stack beneath Soulweaver and the MOCA Library.
The problem
All the intelligence. None of your context.
The difference between a useful agent and a dangerous one is not model size but what it knows about your world. Three failures keep agents from knowing anything at all:
Cold start
Every tool invocation begins contextless. Agents re-derive the same facts about your world, over and over, at your expense.
Session amnesia
Decisions and preferences learned in one conversation evaporate when the session ends. Nothing compounds.
Retrieval unpredictability
Plain vector lookups return close-enough chunks with no sense of who relates to what. Confidence without grounding.
Inside Cortex
From documents to intelligence.
Drop in PDFs, markdown, repos, or whole websites. Cortex extracts the entities and relationships, connects them across document boundaries, and serves the result to every agent you run.
Knowledge graph & GraphRAG
Entities, typed relationships, and community summaries are extracted automatically from your documents — then traversed multi-hop at query time.
Hybrid retrieval
Vector similarity, BM25 keyword matching, and graph traversal in a single query, re-ranked by a cross-encoder. Entity-aware context, not close-enough chunks.
Skills hub
Portable, plain-text SKILL.md capabilities that flow both ways — import community skills or export your graph's know-how to 30+ agent products.
App ecosystem
Web crawler, chat interface, GitHub connector, YouTube importer — every connected app deepens the same knowledge graph.
API-first
60+ REST endpoints with full OpenAPI docs, webhooks, and background task tracking. Collections namespace knowledge per agent, customer, or use case.
Ask AI, with receipts
Source-cited answers with streaming responses and an agentic multi-step mode — for humans and agents alike. Nothing asserted without attribution.
Cortex at the museum
Proven on one of the world's largest digital art collections.
Cortex didn't start as a product. It grew out of the applied research behind the MOCA Library — and the Library already runs on the exact product that's about to be released. Every question you ask it is answered by Cortex, in production, today.
Powers the Library
Every answer in the museum's Library is a Cortex query — collection-scoped search, deep research, and citations over MOCA's institutional memory.
Grounds Soulweaver
Each onboarded collection gets its own Cortex knowledge graph. Holders contribute lore, curators approve it, and every woven soul cites that shared canon.
MOCA flatrate
DeCC0 holders and $MOCA stakers get the Enthusiast tier free — the museum's community runs on the same memory layer it helped build.
Give your agents real memory.
Cortex is the infrastructure behind agents that actually know. Read the docs now — the platform opens soon.