Data flows, identity, and knowledge
Two loops, one triage mechanism, and a knowledge base that grows every day.
Flow C
The Triage Mechanism
Decides whether data updates memory passively or triggers Twin Activation.
Update Knowledge Memory
Low-significance data enriches memory silently.
Trigger Twin Activation
High-significance signals surface as Proposed Actions.
System Data Flows
Two loops. One Spine.
SaaS data → Connectors → Spine → Entity 360 → Twin monitoring
Raw tool data normalized and stored as canonical entities.
AI Session Summaries → Flow C Triage → Knowledge UI validation → Spine update → Refined Twin reasoning
Human interactions validated and stored as structured knowledge.
Three Data Paths
Structured. Unstructured. AI Sessions.
All data becomes unified operational context.
Entity 360 — CRM, billing, support, usage
Every record normalized into one canonical object.
- Account → Contacts → Deals → Tickets → Invoices
- Cross-tool canonical schema
- Automatic relation graphs
Documents, threads, transcripts
PDFs, emails, and conversations parsed as identity data.
- MCP document ingestion
- Semantic NLP extraction
- Communication thread linking
Sessions grounded in Entity 360
Every session links to real canonical records.
- Session-to-entity grounding
- Knowledge extraction pipeline
- Flow C Triage routing
Cross-tool entity matching and dedup
Automatic matching with confidence scoring and human review.
- Multi-signal match scoring
- 0–100 confidence index
- Manual resolution UI
Knowledge Base
Persistent memory. Not stale retrieval.
Zero-Shot Context
Grows daily without model retraining. Operates immediately.
Beyond RAG
Persistent, evolving narrative — not stale snippets.
Entity-Anchored
All knowledge linked to canonical Spine entities.
Truth Layer
Governance by design
Approval Topology
Configurable routing for human verification by role.
Audit Immutability
Append-only logs — every write is traceable.
Confidence Scoring
0–100 scoring before human presentation.
Governance Enforcement
AI cannot write without human approval. Architectural mandate.
Twin Trigger Scoring
From reactive to predictive
Algorithmic threshold that determines when signals warrant action.
The Full Loop
Eight steps. One continuous cycle.
From raw data ingestion to compounding intelligence — every step converges in the Spine.
Ingest
Data flows from 70+ connectors via OAuth, MCP, or webhook.
Normalize
8-stage pipeline resolves entities and deduplicates records.
Detect
Cross-system patterns and anomalies surfaced automatically.
Analyze
Twin scores signals with confidence and source attribution.
Surface
Morning Brief and Entity 360 present ranked action items.
Approve
Govern Gate routes actions by impact tier for human review.
Execute
Approved actions write back to source systems idempotently.
Learn
Every decision enriches the knowledge base. No retraining.