A knowledge engine that says nothing when it isn’t sure.
While building a knowledge graph for legal and drilling documents, we ended up with something that mimics how a brain reads, remembers, and checks itself. It reads everything you’ve written down, connects what it finds, verifies every fact against the source - and refuses to answer when the evidence isn’t there.
Most systems retrieve, generate, and hand you the answer.
We benchmarked six knowledge-graph frameworks - LightRAG, HippoRAG, PathRAG, OG-RAG, Graphify and PageIndex - and found the same gap in all of them: they retrieve and generate, then hand you the answer with no verification. For high-stakes documents like contracts and drilling reports, that is an unacceptable risk. The Cognitive Index was built to close that gap.
Nine parts, borrowed from how a brain works.
Personalized PageRank
Relevance spreads outward from what you asked, the way a memory pulls related memories with it.
Leiden communities
The graph organizes itself into specialized regions, so related knowledge sits together.
RAPTOR
Detail is summarized into higher-level memory, so the system can reason at the right altitude.
Fast fact filtering
A quick pass recognizes what’s relevant before the heavy reasoning begins.
Temporal document guards
Facts stay anchored to when they were true, so old and new evidence never blur together.
Query decomposition
Complex questions are broken into a plan of smaller, answerable steps.
Critics pipeline
A pipeline of critics reviews candidate answers before anything reaches you.
Numeric verification
Every figure in an answer is checked back against the evidence it came from.
Consensus-weighted edges
Connections backed by more agreement carry more weight - the graph learns what to trust.
It shapes itself to your data - then verifies.
Rather than force your documents into a fixed schema, the Cognitive Index discovers the entity types and relationships specific to your domain, then deletes the modules it doesn’t need. It handles both prose and structured data, and - crucially - it grounds and self-checks every answer instead of trusting the model’s first response.
Slower on purpose, where being wrong is expensive.
Full-mode queries take minutes and real compute - because accuracy, source attribution and self-verification come first. Faster competitors win on speed and cost for simpler questions. The Cognitive Index is built for the decisions where a confident wrong answer is the expensive outcome.
Built for high-stakes documents.
Legal & compliance
Contracts, circulars and obligations where a missed clause is a real liability.
Oil & gas
Drilling plans, daily reports and close-outs across years of operations.
Financial due diligence
Deal books and statements where every number must trace to its source.
Healthcare
Dense clinical and regulatory documents where accuracy is non-negotiable.
Risk & operations
Recurring hazards hiding across thousands of pages, linked back to evidence.
Knowledge capture
Hard-won expertise, kept searchable before the people who hold it move on.
See the engine work on your own documents.
Spend 30–45 minutes with our expert to see the Cognitive Index applied to your archive.