6 hrs → 7 sec
Document question resolution at a professional-services firm after deploying document intelligence
The day your company forgot.
Company Brain is the memory that never resigns — everything your organization knows, structured, governed, and answerable in seconds. With citations.
What is a company brain?
A company brain is a governed enterprise knowledge layer: documents, data, and institutional expertise ingested into a knowledge graph with a semantic layer, so both people and AI agents get accurate, cited answers grounded in the organization's own truth — not the internet's.
The Context Tax is real, recurring, and entirely optional.
35,000+
Operational documents under one Brain at a global aviation operator
60×
Faster resolution for the aviation operator's dispatch and ops teams — mid-flight, every time
Not a search box. A memory with structure, meaning, and receipts.
- Structure
Knowledge graph & ontology
Entities and relationships extracted from everything ingested — grounded against the Enterprise Twin, so “the Atlanta plant” is the same node everywhere.
- Meaning
Semantic layer
“Net revenue” and “active client” defined once, used identically by every person and agent. The antidote to twelve departments, twelve definitions.
- Answers
Enterprise RAG with citations
Hybrid retrieval — vector, keyword, graph traversal — with reranking and citation-first answers. Every claim traceable to its source document.
- Ingestion
Document intelligence
Parsing, OCR, and table extraction that turned a 6-hour task into seconds at a professional-services firm. Contracts, runbooks, drawings, emails.
- Continuity
Persistent memory (Engram)
Agents and teams accumulate episodic memory over time — decisions made, exceptions granted, context earned — instead of starting cold every session.
- Trust
Governed by design
Access scopes, PII masking, and hallucination checks from the Governance plane apply to every query. The Brain only answers what the asker is allowed to know.
What happens when you ask a question.
Answers return inline [n] citations with their sources.
You ask a question
Answerinline [n] citations
01Semantic search
Semantic search over embeddings — run together with lexical keyword search.
Over embeddingsRun together with lexical
02Lexical search
Lexical keyword search — run together with semantic search over embeddings.
Keyword searchRun together with semantic
03Rank fusion
Results fused by reciprocal-rank fusion.
Reciprocal-rank fusion
04Reranker
An optional reranker reorders for relevance.
OptionalReorders for relevance
05Graph expansion
The knowledge graph expands to connected context.
Knowledge graphConnected context
Every step filtered by the asker's source permissions.
Memory that matures.
Engram memory moves through four stages — with per-stage decay and promotion.
01
Ephemeral
02
Working
03
Consolidated
04
Crystallized
Contradiction detection between memories · Per-user GDPR forget
The semantic layer and the knowledge graph.
Semantic layer
Canonical business terms and metrics are injected into answers, so “net revenue” means one thing everywhere.
net revenuecanonical term
Formula
Grain
Synonyms
Owner
Knowledge graph
Typed entities and typed relations — grounded against the Enterprise Twin.
Typed entities
People · Teams · Systems · Products · Policies · Metrics · Documents · Concepts
Typed relations
owns · depends_on · reports_to · references · part_of · responsible_for
The knowledge was always somewhere. The difference is that now “somewhere” answers in seven seconds, cites its sources, and doesn't take vacations.
The Context Tax: what not knowing costs you every day
The white paper that names the invisible line item: how repeated questions, tribal knowledge, and key-person risk compound into a seven-figure annual cost — and the three-phase playbook for eliminating it.
Asked before every Brain deployment
Search finds documents; a RAG chatbot summarizes them. A company brain adds structure (knowledge graph), shared meaning (semantic layer), memory over time (Engram), and governance on every answer — which is what makes it reliable enough for agents to act on, not just people to read.
The Brain inherits and enforces source-system permissions, layered with SphereIQ policies. A person or agent only retrieves what they’re entitled to see — enforced at query time by the Governance plane, with PII masking where policy requires it.
First collections are typically live in weeks: connect the highest-value sources via SphereIQ Connect, ingest, and validate answers with the owning team. Value compounds from there as coverage and the semantic layer grow.
A memory is powerful. A memory that can see the whole enterprise is a different thing entirely.
Chapter 03 is the one nobody else can write.