End-State Architecture
The complete SpineWorkspace architecture — one workspace, all connected software, shared context, human + AI working together
One Workspace. All Connected Software. Shared Context. Human + AI Working Together.
This is the canonical end-to-end architecture reference. Every other architecture doc is a deep-dive into one section of this page.
The Flow
Every piece of work in SpineWorkspace follows one continuous loop:
Business Context → Work → Communication → Decisions → Execution → Reconciliation → Memory → Next Work
Nothing is isolated. Every action feeds back into organizational memory. Every decision is traceable. Every outcome creates the context for the next cycle.
Architecture Layers
The architecture is composed of 10 interconnected layers, organized into three horizontal bands:
Top Band — User Surfaces
| Layer | Purpose | Docs |
|---|---|---|
| Users & Roles | Who uses the system | Roles & Agents |
| Human Workbench | One workspace for all human work | Human Workbench |
| Communications Layer | All conversations in one place | Communications |
| Twin Workbench | AI workspace for reasoning, planning and execution | Twin Workbench |
Middle Band — Platform Core
| Layer | Purpose | Docs |
|---|---|---|
| Source Systems | External applications and tools | Bridge Layer |
| Bridge Layer | Connectivity and integration | Bridge Layer |
| Spine Fabric | Canonical business memory and operational continuity | Spine Schema |
| ADX Orchestrator | Agent runtime and orchestration | ADX Orchestrator |
| Governance Layer | Policy, authorization, approval | Governance |
| Execution Fabric | How work gets done | Execution Fabric |
| Target Systems | External systems being acted upon | Bridge Layer |
Bottom Band — Continuity
| Layer | Purpose | Docs |
|---|---|---|
| Reconciliation & Continuity Pipeline | Closes the loop on every action | Reconciliation Pipeline |
| Data Governance & Security | Cross-cutting concerns across all stores | Data Governance |
Users & Roles
SpineWorkspace serves every role in the organization. Each role gets a domain-specific projection of the workbench.
Human Roles
| Role | Primary Focus |
|---|---|
| Executive Leader | Strategic KPIs, organizational health, decisions |
| Sales Manager | Pipeline, opportunities, forecasts |
| Account Manager | Customer health, renewals, expansion |
| Finance Analyst | Revenue, billing, collections |
| Support Lead | Tickets, SLAs, escalations |
| Any Role / Customer | Role-specific work surfaces |
AI Agents
| Agent Type | Scope | Behavior |
|---|---|---|
| AI Twin | Per-user, role-scoped | Reasons, plans, suggests — never acts without approval |
| AI Coworker | Context companion | Observes active work, gathers context, supports communication |
| Domain Agents | Sales, Finance, Support... | Specialized reasoning within a business domain |
| Specialist Agents | Research, Analysis... | Deep-dive capabilities triggered on demand |
| System Agents | Platform Services | Infrastructure, scheduling, monitoring |
See Roles & Agents for full definitions.
Human Workbench
One Workspace for All Human Work.
The Human Workbench is the single surface where every person does their job. No context switching. No app juggling.
Navigation
| Section | Purpose |
|---|---|
| My Work | Personal work items and priorities |
| My Queue | Ordered work waiting for attention |
| Today | Today's focus and timeline |
| Alerts | Notifications requiring attention |
| Approvals | Items awaiting human approval |
| Decisions | Decisions requiring human judgment |
| Reports & KPIs | Analytics and performance metrics |
| Settings | Configuration and preferences |
Work Surface Tabs
| Tab | Purpose |
|---|---|
| Orientation | Context and situational awareness |
| Work | Active work items and tasks |
| Communications | All conversations and messages |
| Organizations | Company and relationship context |
| Knowledge / Memory | Organizational knowledge base |
| Domain Work Surfaces | Role-specific deep-dive views |
Domain Work Surfaces
Each role gets a tailored view of the work surface:
| Domain | Focus |
|---|---|
| Sales | Pipeline, Opportunities, Forecasts |
| Account Success | Health, Renewals, Expansion |
| Finance | Revenue, Billing, Collections |
| Support | Tickets, SLAs, Escalations |
| Operations | Projects, Tasks, Workflows |
| Custom Views | Role / Domain Specific Projections |
Business KPI / Health Overview
The workbench surface key metrics at a glance:
| Metric | Value | Trend |
|---|---|---|
| Revenue | $8.4M | +12% |
| Pipeline | $27.6M | +8% |
| Revenue at Risk | $1.2M | +15% |
| Customer Health | 76 | +4 pts |
| Open Work | 128 | +2 |
| SLA Risk | 5 | View all KPIs → |
See Human Workbench for complete specification.
Twin Workbench
AI Workspace for Reasoning, Planning and Execution.
The Twin Workbench is the AI counterpart to the Human Workbench. It reasons about work, proposes actions, and executes under governance.
Tabs
| Tab | Purpose |
|---|---|
| My Twin View | Twin's current understanding and state |
| Active Work | Work items the Twin is reasoning about |
| Reasoning & Plan | The Twin's analysis and proposed plan |
| Tools & Capabilities | Available AI tools and integrations |
| Memory | What the Twin knows and remembers |
Context & Understanding
The Twin maintains a continuous understanding of the current situation:
- Current State — What is happening right now
- Relevant Context — Related information from the Spine
- Evidence — Data supporting or contradicting actions
- Related Work — Connected work items and dependencies
- Risks & Drivers — What could go wrong, what matters most
Suggested Actions (OODA Grammar)
The Twin follows the OODA loop for every work item:
- Understand Context → Orient to the situation
- Collect Evidence → Gather supporting data
- Analyze → Reason about options
- Recommend → Propose a course of action
- Plan → Detail the execution steps
- Awaiting Approval → Human decides
- Execute → Act on the approved plan
- Reconcile → Verify the outcome
- Complete → Close the loop
AI Tools & Capabilities
| Tool | Purpose |
|---|---|
| Research | Deep investigation and information gathering |
| Data Analysis | Quantitative analysis and pattern detection |
| Prediction | Forecasting and trend analysis |
| Drafting | Content creation and communication |
| Workflow | Automated process execution |
| Automation | Rule-based task completion |
| Integration | External system interaction |
Coworker (Context Companion)
The Coworker is a lighter-weight AI participant that:
- Observes active work without interfering
- Gathers relevant context from the Spine
- Retrieves information on demand
- Supports communication between humans and the Twin
Focus Areas: Active Work Context, Relevant Information, Communication Support, Context Alignment.
See Twin Workbench for complete specification.
Communications Layer
All Conversations in One Place.
The Communications Layer unifies every channel of communication into a single surface:
| Channel | Description |
|---|---|
| Team Chat | Internal team messaging |
| Customer Communications | External customer-facing messages |
| Internal Updates | Status updates and announcements |
| Email integration and management | |
| Meetings | Meeting notes, action items, follow-ups |
| Announcements | Organization-wide communications |
| AI / Human Hand-offs | Transitions between AI and human actors |
The Communications Layer feeds directly into the Spine as events and evidence. Every communication is searchable, traceable, and linked to the work it relates to.
See Communications Layer for complete specification.
Spine Fabric
Canonical Business Memory & Operational Continuity.
The Spine Fabric is the heart of SpineWorkspace — the single source of truth that all other layers read from and write to.
Core Domains
| Domain | Purpose |
|---|---|
| Business Context | Goals, Objectives, Needs, Standards |
| Work Graph | Work, Tasks, Activities, Dependencies |
| Entities | Accounts, People, Products, Assets... |
| Events & Evidence | Activities, Changes, Interactions, Signals |
| Decisions | Choices, Rationales, Approvals |
| State & Status | Current, History, Attribution |
| Provenance | Source, Time, Attribution |
| Organizational Memory | Knowledge, Insights, Learnings |
Spine Stores
The Spine operates across three storage tiers, each with distinct governance:
| Store | Technology | Contents |
|---|---|---|
| Canonical Spine | Neon / Postgres | Authoritative truth, long-term memory, governed writes |
| Operational Spine / Context Store | D1 / Durable Objects / Cache | Working context, real-time state, short-term working sets |
| Edge / Session State | Workers / Cache | Ephemeral state, per-session / per-device, fast access |
See Spine Schema for complete data model.
Bridge Layer
Connectivity & Integration.
The Bridge Layer connects SpineWorkspace to every external system:
Source Systems
Salesforce, HubSpot, Dynamics 365, NetSuite, QuickBooks, ServiceNow, Zendesk, Slack / Teams, Google Workspace, Microsoft 365, Databases / APIs, Files / Events / Webhooks, Any System.
Bridge Components
| Component | Purpose |
|---|---|
| Connectors | Pre-built system integrations |
| Adapters | Protocol and format translation |
| APIs | Direct API connections |
| Webhooks | Event-driven ingestion |
| Event Streaming | Real-time data feeds |
| File / DB Sync | Batch and incremental synchronization |
| Transformers | Data normalization and enrichment |
| Protocol Handlers | Protocol-specific implementations |
Target Systems
The same systems listed as sources can also be targets — the Bridge Layer is bidirectional.
See Bridge Layer for complete specification.
ADX Orchestrator & Agent Runtime
Orchestration Engine + Agent Runtime + Workflow Engine.
The ADX Orchestrator is the brain that coordinates all agent activity:
| Component | Purpose |
|---|---|
| Orchestration Engine | Coordinates multi-step agent workflows |
| Agent Runtime | Executes individual agent actions |
| Workflow Engine | Manages complex business processes |
| Task Decomposition | Breaks work into executable units |
| Context Assembly | Builds the context window for agents |
| Agent Hand-off (A2CP) | Transfers work between agents |
| State Management | Tracks workflow and agent state |
| Scheduling | Time-based and event-based triggers |
| Retries | Automatic retry on failure |
| Monitoring | Observability and alerting |
See ADX Orchestrator for complete specification.
Governance Layer
May This Be Done?
The Governance Layer is a hard gate between intent and execution. Nothing executes without passing through governance.
| Component | Purpose |
|---|---|
| Policy Evaluation | Checks against organizational policies |
| Authorization | Verifies permissions and roles |
| Approval Workflow | Routes to human approvers when required |
| Limits & Guardrails | Enforces spending, rate, and scope limits |
| Risk & Compliance | Assesses risk and compliance requirements |
| Audit & Accountability | Records who did what, when, and why |
Governance Rules
- No execution token = no execution
- Every action is auditable
- Human approval required for high-risk actions
- Twin never acts without explicit approval
See Governance for complete specification.
Execution Fabric
How It Gets Done.
The Execution Fabric is the set of capabilities that actually perform work:
| Component | Purpose |
|---|---|
| APIs & Services | Direct API calls to external systems |
| MCP / Tools | Model Context Protocol tool invocations |
| Workflows | Multi-step process execution |
| Hermes | Execution service for complex actions |
| Functions / Workers | Serverless compute execution |
| External Integrations | Third-party service calls |
See Execution Fabric for complete specification.
Reconciliation & Continuity Pipeline
Closes the Loop on Every Action.
The Reconciliation Pipeline ensures every action is verified and feeds back into memory:
Execution Receipt → Re-ingestion → Normalization & Entity Resolution → State Update → Reconciliation → Discrepancy Handling → Memory Promotion → Next Work
| Step | Purpose |
|---|---|
| Execution Receipt | Confirm Acknowledgement |
| Re-ingestion | New Data / Events |
| Normalization & Entity Resolution | Resolve entities across stores |
| State Update | Spine Update |
| Reconciliation | Intended vs. Actual |
| Discrepancy Handling | Retry / Compensate |
| Memory Promotion | Knowledge / Learning |
| Next Work | Created / Updated |
Loop State (Stored in Spine)
Every work item progresses through these states:
Identified → Contextualizing → Defined → Allocated → Deciding → Governing → Executing → Observing → Re-ingesting → Reconciling → Resolved → Learning → Continuing → Next Work
See Reconciliation Pipeline for complete specification.
Data Governance & Security
Across All Spine Stores.
Cross-cutting concerns that apply to every layer:
| Concern | Scope |
|---|---|
| Tenants | Multi-tenant isolation |
| Identity | Authentication and identity |
| Roles | Role-based access control |
| Permissions | Fine-grained permission model |
| Policies | Organizational policy enforcement |
| Purpose | Purpose limitation for data use |
| Consent | Consent management |
| Data Classification | Sensitivity and handling labels |
| Audit | Immutable audit trail |
| Retention | Data lifecycle management |
See Data Governance & Security for complete specification.
Outcomes
SpineWorkspace delivers:
- Faster execution — Work moves from intent to outcome without friction
- Lower risk — Every action is governed, audited, and reversible
- Higher quality decisions — AI reasoning backed by complete context
- Audit ready — Full traceability from decision to outcome
- Connected teams — One surface, no silos, shared context
- Organizational intelligence — Every cycle builds institutional knowledge