One substrate · six shapes · live instance
Everything below runs against a real OriginChain instance.
Nothing is canned.
OCCI — OriginChain Context Intelligence. One banking dataset seeded as real tables; SQL, graph, vector, full-text, and natural-language queries all hit the same substrate, and the context-intelligence layer governs what any of it is allowed to tell an LLM.
Standard SQL over the banking dataset — joins, GROUP BY, self-joins — one substrate, no second engine.
Ownership and transaction edges traversed in place — BFS to PageRank, a nine-motif laundering library, and LP-bounded pattern planning.
Three index kinds across the scale curve, plus multimodal search — voiceprints, KYC images, video, behavioral embeddings — ending in retrieved chunks feeding a live LLM answer.
400 support tickets, tokenized at write time — boolean, phrase, and BM25-ranked search with scores you can see.
Ask the database in plain English. Deterministic grammar first, LLM fallback second, cached plans after that — with the plan visible.
The full assistant surface: chat with the governed pipeline, watch the tool trace and before/after context diff live, per-call cost and audit trail included.
The six-screen walkthrough: deterministic relevance, role, PII, and compression decisions before the model runs — plus governed live chat.
examples run in mock mode until the instance credentials land — each response is badged live / mock