IntegrateWise — Final Canonical Model
Status: SUPERSEDED PLATFORM SYNTHESIS. This file does not supersede the
frozen architecture. Any conflicting product, Twin, schema, storage or
execution claims are historical. Current authority starts at
../architecture/ARCHITECTURE_LOCK.md.
Original status: CANONICAL — Supersedes all prior decks, specs, and architectural documents.
Owner: IntegrateWise Platform
Last Verified: 2026-07-25
UX Reference: Workbench should be similar to UI Bakery style
1. Final definition
IntegrateWise is operational continuity infrastructure that connects an organisation's tools, AI systems, data and work through one governed Spine Network, then projects the right entities, context, memory and capabilities into a role-aware User Workbench.
The user experiences:
ONE CONNECTED ECOSYSTEM
+
ONE USER WORKBENCH
+
ONE TWIN
The platform underneath maintains:
SPINE
MEMORY
PROJECTION
GOVERNANCE
CAPABILITIES
EXECUTION
CONTINUITY
2. The complete model
EXTERNAL ECOSYSTEM
SaaS Tools · Databases · Files · MCP · AI Providers · Internal APIs
│
▼
ONE AUTH-TO-ECOSYSTEM
│
▼
INTEGRATION MANAGER
API Connectors · MCP · AI Connectors · API Wrappers · Webhooks
│
▼
GATEWAY
Ingress · TenantGate · Authentication
Scope · Routing · Trace Context
│
▼
LOADER RUNTIME
┌─────────────────┬─────────────────┬─────────────────┬─────────────────┐
│ │ │ │ │
▼ ▼ ▼ ▼ ▼
FLOW A FLOW B FLOW C FLOW D INGESTION
Structured Unstructured AI Context Platform Queues
Operational Conversational Cognitive Binding Per Flow
3. Flow A — Structured operational data
Flow A owns deterministic operational records:
Entities · Tuples · Fields · Relationships · Metrics · Events · Statuses · Ownership · Lifecycle state
Typical sources:
CRM · Support · Billing · Finance · Projects · Tasks · Product · HR · ITSM · Sales · Marketing · Engineering
Flow A is the primary structured substrate projected into Workbench tables, entity views, boards, queues and timelines.
4. Flow B — Unstructured context
Flow B owns communication, documents and semantic context:
Conversations · Threads · Messages · Documents · Attachments · Transcripts · Chunks · Embeddings · Entity links · Knowledge references
Flow B enriches Flow A:
Account entity
+
Support conversation
+
Email history
+
QBR document
+
Call transcript
=
Account continuity
5. Flow C — AI and cognitive continuity
Flow C owns AI-originated and AI-mediated material:
Twin sessions · External AI chats · Model outputs · Plans · Proposals · Tool calls · Generated artefacts · Candidate memory/knowledge/decisions · Agent outcomes
Flow C follows:
AI context → Brainstorming/session layer → Triage → Candidate → Evidence → Governance → Promote or reject
Flow C cannot directly mutate operational truth.
6. Flow D — Platform binding
Flow D is the stable platform-binding plane:
Identity · Tenant · Workspace · Membership · Role · Department · Industry · Permissions · Configuration · Connections · Credential references · Capabilities · Policies · Entitlements · Runtime bindings
Flow D binds Flow A, B and C to the correct trust, user, workspace and policy context.
7. Gateway
Ingress · TenantGate · Authentication · Scope · Routing · Trace Context
8. Normalizer and Triage (separate by design)
Deterministic Normalizer
Analyze → Classify → Filter → Refine → Extract → Validate → Sanity Check → Resolve identity → Sectorize
Triage Runtime and Promotion
Classify AI context → Extract candidate memory/knowledge → Attach evidence → Score confidence → Route review → Promote/reject
Platform Binding Validation
Identity · Tenant · Workspace · Role · Department · Industry · Permissions · Configuration
9. Core Spine — Stable Schema
Operational Truth
Entities · Tuples · Relationships · Events · Metrics · Lifecycle State · Commitments · Decisions
Knowledge & Evidence
Documents · Chunks · Embeddings · Source Links · Intelligence Evidence · Semantic Context · Provenance · Retrieval References
Memory & Continuity
User Memory · Work Memory · Org Memory · Entity Memory · Decision Memory · Outcome Memory · Continuity State · Agent Memory
10. Stable Core, Adaptive Projection
The Core Spine schema remains stable. The user-facing schema varies through projection.
Stable Core Spine
+
Projection Metadata
+
Active User Context
+
Policy
=
Adaptive User Workbench
Projection dimensions
Role × Department × Industry × Connector Category × Active Context × User and Tenant Policy × Available Capabilities × Data Freshness
11. Adaptive User Workbench
Tables · Entity Views · Pages · Boards · Queues · Timelines · Calendars · Documents · Approvals · Operational Canvases · Contextual Capabilities
UX Rule: Workbench should be similar to UI Bakery style
Normal state = Workbench-led
User clicks Ask / Assign / Proposal / Signal = Twin detail opens contextually
User clicks Chat AI = Conversational interface opens
User exits = Workbench returns to primary focus
Twin is a contextual next-panel, drawer or overlay attached to the active entity/task/decision/signal/document/queue/proposal/approval.
Chat AI — Flow C Entry Point
The Chat AI is a complete conversational interface that:
- Maps directly to the Spine — reads and writes through governed paths
- Brainstorming layer — free-form ideation, research, planning
- Triage the Twin — candidate memory, knowledge, and decisions flow from Chat AI through Triage into the Spine
- Evidence-backed — every AI output shows its sources and confidence
- Dual-mode:
- Exploration mode — brainstorm, research, ask questions
- Action mode — propose tasks, create entities, generate plans
The Chat AI is NOT a replacement for the Twin. They serve different purposes:
Chat AI = Exploratory intelligence (Flow C entry point)
Twin = Contextual intelligence (per-entity activation)
Chat AI triage path:
Chat AI conversation
→ Candidate memory/knowledge/decisions
→ Triage Runtime
→ Evidence attached
→ Governance review
→ Promote to Spine or reject
The user can move seamlessly between Chat AI and Twin:
- Chat AI generates a proposal → Twin surfaces it in context
- Twin identifies a gap → User opens Chat AI to explore
- Chat AI creates a task → Twin shows it in the entity detail
12. Twin
Observe · Orient · Reason · Explain · Plan · Propose · Delegate · Monitor
The Twin must not occupy a permanently open sidebar by default.
13. Capability Fabric
Tool-to-Tool · Memory Fetch · MCP · Agent-to-Agent · Local · API · Workflow · Browser · Human Task
14. Governance
Policy · Authority · Evidence · Risk · Scope → Allow · Deny · Require Approval
15. Hermes (Execution Coordinator)
Bind approved plan · Select activated route · Coordinate · Retry · Monitor · Collect evidence · Reconcile
16. Execution Providers
API · MCP · Agent · Browser · Human · Workflow
17. Provider Outcome
Native acknowledgement · ID · Status · Evidence · Error · Result payload
18. Sync and Reconciliation
Re-ingest · Normalize · Compare · Verify · Resolve conflict · Close loop
19. Final continuity loop
CONNECT → AUTHENTICATE → INGEST → LOAD → NORMALIZE / TRIAGE → GOVERNED WRITE → SPINE → PROJECT → WORK → ASK / ASSIGN → PROPOSE → GOVERN → EXECUTE → ACKNOWLEDGE → RE-INGEST → VERIFY → RECONCILE → REHYDRATE → CONTINUE
20. Final freeze statement
IntegrateWise uses One Auth and the Integration Manager to connect the external ecosystem, Gateway and Loader Runtime to route information through Flow A, Flow B, Flow C and Flow D, deterministic Normalization and governed Triage to write into one stable Core Spine, and the Projection Engine to compose an Adaptive User Workbench based on role, department, industry, connector category, active context and policy. The Twin appears contextually to reason and propose; Capability Fabric resolves execution, Governance controls authority, Hermes coordinates the approved work, and Sync and Reconciliation verify the provider outcome before the Spine rehydrates the Workbench.
21. What is no longer canonical
- 16 named agents as the primary product architecture
- 21 Hubs as separate applications
- Atlas Spine as a separate older model
- Different physical schemas per role or connector
- AI directly writing truth
- Automatic execution based only on model confidence
- OpenWebUI as the Twin runtime
- Nango as the identity of the whole Integration Manager
- Dashboards as the Workbench
- Static departmental applications
- Direct database access from Workbench components
- Direct execution from Twin
- Connector installation permanently adding arbitrary columns
- AI chat transcripts directly becoming organisational memory
Agents, hubs, providers and infrastructure remain implementation options underneath the stable canon.