The data and AI platform
nothing else combines.
Nobody else ships this stack. Open-source knowledge graph that combines vector, graph, document, immutable history and time travel — with a Memory, Heart and Conscience layer running on top. All self-hostable. All your data.
The pieces have existed separately for years. Pinecone has vectors. Neo4j has graphs. Datomic has time travel. OpenAI has a "memory" feature. Nobody has the whole stack, open-core, and on the hardware you already own. Until now.
01The data foundation: six-for-six
Every vector database forces you into a stack. Pinecone for vectors, Neo4j for graphs, Postgres for metadata, Datomic if you need time travel, duct tape to glue it all. Brainy + Cor collapses the stack into one engine.
| Product | Vector | Graph | Document | Immutable history | Time travel | Open-core |
|---|---|---|---|---|---|---|
| Pinecone | ✓ | ✗ | ✗ | ✗ | ✗ | ~ cloud-only |
| Qdrant | ✓ | ✗ | ~ partial | ✗ | ✗ | ~ partial |
| Weaviate | ✓ | ~ partial | ~ partial | ✗ | ✗ | ~ partial |
| FAISS | ✓ | ✗ | ✗ | ✗ | ✗ | ✓ |
| Neo4j | ✗ | ✓ | ~ partial | ✗ | ~ partial | ~ commercial |
| Datomic | ✗ | ~ partial | ✓ | ✓ | ✓ | ~ JVM, closed |
| TerminusDB | ✗ | ✓ | ✓ | ✓ | ~ partial | ~ partial |
| Brainy 8.0 + Cor 3.0 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
No other open-source data platform combines all six. Datomic comes closest — but it's closed JVM, no vectors. TerminusDB gets close — but it's partial on time travel and partial on open-core. Pinecone is cloud-only. Weaviate and Qdrant need a cluster. Neo4j is commercially licensed and has no vectors. Brainy + Cor ships all six. Open-core. Self-hostable. 10B+ on a laptop, 100B+ on a server.
02AI infrastructure, not AI features
Every AI product treats memory, empathy, and values as features on a chat surface — a memory pin, a system prompt, a vendor-set refusal filter. Soulcraft treats them as infrastructure: three pillars sitting beside the data layer, every product reading from the same Memory, Heart, and Conscience.
Persistent memory across sessions
Pinned notes, RAG lookups, or rolling summaries — facts stored, nothing learned
Memory consolidates every night; builds dossiers, abstracts patterns, prunes noise
Emotional intelligence
System-prompt tweaks ("be friendly"). Same flat tone for every kind of moment
Heart reads tone, weights memories by emotional salience, shapes responses pre-send
Values alignment
Vendor-set safety filters that refuse a fixed list — your values are not in the loop
Conscience holds your stated principles + observed judgment style; checks every response
Cross-product continuity
Each AI surface starts cold — no shared memory between coding and customer-reply tools
One Memory, one Heart, one Conscience across Workshop, Venue, Academy, and every Soulcraft surface
Open-core, self-hostable
Vendor cloud only. Your conversations and values live on their servers, by their rules
Memory runs on Brainy + Cor on your hardware. Stop paying and your data still works
03Open-core, end to end
The data layer is MIT (Brainy). The AI services run on your hardware. You can stop paying for any commercial piece and the system keeps running — just slower at the data layer, with fewer AI capabilities at the top. There is no managed-service tax anywhere in the stack.
Stop paying and your data still works. Drop Cor — Brainy keeps serving queries, slower. Drop the AI services — Workshop, Venue, and Academy still run on the data layer. No piece of the stack holds the next one hostage. That's not a marketing promise; it's the architecture.
04One database per user — not one cloud for everyone
Managed AI pools every customer's data into one multi-tenant system. Soulcraft inverts it: every user gets their own Brainy database. Isolation isn't a policy promise on top of a shared cloud — it's the architecture underneath.
Your memories, emotional context, and values live in a database that's yours alone. Nothing mixes with anyone else's; nothing trains someone else's model.
Download the whole thing as a file, anytime. It's a standard Brainy database — the same open format that runs on a laptop. No export queue, no lock-in.
Erase the file and it's genuinely gone — not a 30-day "deletion request." There's exactly one place your data lives, so removing it is real and immediate.
Multi-tenant by exception, not by default. Most AI platforms can't offer this because their economics depend on one shared index. Brainy + Cor runs a full database per user on commodity hardware — so isolation costs nothing and privacy is structural.
05What this combination unlocks
Each capability matters on its own. Combined, they enable workloads that simply can't be built on managed-only stacks.
A knowledge graph that learns from itself
Vector recall + graph traversal + temporal history in one query. The agent that needs your project context, the customer record from last year, and the conversation about it — all from one call, not three integrations.
Time-travel debugging for AI behavior
Immutable history + Datomic-style asOf queries mean you can ask "what did the AI know on Tuesday when it gave that answer?" — the kind of audit story that managed services can't deliver.
AI that gets smarter every night, on your hardware
Memory's overnight consolidation runs against your own data, on your own machines, paying your own electricity bill — not a per-token meter. The 10B-on-a-laptop architecture makes this economic.
One values register across every product
What you decided in Workshop applies when Venue drafts a customer reply. What you said you stood for last month catches drift today. Conscience is values infrastructure, not a per-product setting.
The whole stack. One platform.
Brainy 8.0 + Cor 3.0 ship the data foundation. Memory, Heart, and Conscience arrive in Q3 2026. Workshop, Venue, and Academy launch alongside. Join the waitlist for any product and we'll keep you in the loop.
Architectural claims based on Brainy 8.0 + Cor 3.0 design specs. Specific recall and latency numbers ship with the Cor verification report alongside the GA release.