system-prompts-and-models-of-ai-tools/salesflow-saas/docs/ai-operating-model.md

Dealix AI Operating Model — Five-Plane Architecture

Parent: MASTER_OPERATING_PROMPT.md
Version: 1.0 | Status: Canonical
Tracks: All six tracks

---

Overview

Dealix separates concerns into five architectural planes. Each plane has a distinct responsibility, clear boundaries, and explicit contracts with adjacent planes.

┌─────────────────────────────────────────────────┐
│              DECISION PLANE                      │
│   Strategy · Forecasting · Memos · Evidence      │
├─────────────────────────────────────────────────┤
│              EXECUTION PLANE                     │
│   OpenClaw · Durable Flows · Agents · Celery     │
├─────────────────────────────────────────────────┤
│              TRUST PLANE                         │
│   Policy Gates · Approvals · Audit · Compliance  │
├─────────────────────────────────────────────────┤
│              DATA PLANE                          │
│   PostgreSQL · pgvector · Redis · Events · RAG   │
├─────────────────────────────────────────────────┤
│              OPERATING PLANE                     │
│   CI/CD · Monitoring · Self-Improvement · Flags  │
└─────────────────────────────────────────────────┘

---

1. Decision Plane

Purpose: Where strategic decisions are made, forecasts generated, and executive memos assembled.

Current State

Component	File	Status
Executive ROI Service	services/executive_roi_service.py	Live (basic)
Analytics Service	services/analytics_service.py	Live
Management Summary Agent	ai-agents/prompts/management-summary-agent.md	Live
Revenue Attribution Agent	ai-agents/prompts/revenue-attribution-agent.md	Live
Predictive Revenue	services/predictive_revenue_service.py	Live
Strategic Simulator	services/strategic_deals/strategic_simulator.py	Live
ROI Engine	services/strategic_deals/roi_engine.py	Live

Target State

Component	Status
Executive Room (full aggregation)	Building
Evidence Pack Assembly	Building
Actual vs Forecast Control Center	Building
Contradiction-aware decisioning	Building
Board Pack Generator	Planned

Structured Outputs

All Decision Plane outputs must be structured:

• LeadScoreCard — qualification score + signals + recommendation
• QualificationMemo — deal qualification with evidence
• ProposalPack — pricing + terms + value proposition
• ExecutiveSnapshot — KPIs + risks + pending decisions
• EvidencePack — assembled proof for audit/board review
• ForecastVariance — actual vs forecast with root causes

---

2. Execution Plane

Purpose: Where work gets done. Durable, checkpointed, retriable workflows.

Current State

Component	File	Status
OpenClaw Gateway	openclaw/gateway.py	Live
Durable Task Flow	openclaw/durable_flow.py	Live
Task Router	openclaw/task_router.py	Live
Policy Engine	openclaw/policy.py	Live
Approval Bridge	openclaw/approval_bridge.py	Live
Observability Bridge	openclaw/observability_bridge.py	Live
Hooks	openclaw/hooks.py	Live
Canary Context	openclaw/canary_context.py	Live
Plugins (5)	openclaw/plugins/	Live
Agent Executor	services/agents/	Live
Celery Workers	workers/	Live
Sequence Engine	services/sequence_engine.py	Live

Execution Flow

Request → OpenClaw Gateway
  → Policy Gate (policy.py: A/B/C classification)
  → Observability (start run, trace)
  → Approval Bridge (if Class B: check approval_token)
  → Canary Context (if canary enforcement: tenant check)
  → Task Router (dispatch to registered handler)
  → Durable Flow (checkpoint state)
  → Agent Executor / Celery Task
  → Action Handler (DB write, message send, etc.)
  → Observability (finish run)

Target State

Component	Status
Temporal for long-running workflows	Watch
Compensation policies (rollback)	Planned
Idempotency keys for all writes	Planned
Dead letter queue with alerting	Planned

---

3. Trust Plane

Purpose: Where governance is enforced. No sensitive action bypasses this plane.

Current State

Component	File	Status
Policy Classes (A/B/C)	openclaw/policy.py	Live
Approval Bridge	openclaw/approval_bridge.py	Live
Trust Score Service	services/trust_score_service.py	Live
Security Gate	services/security_gate.py	Live
Shannon Security	services/shannon_security.py	Live
PDPL Consent Manager	services/pdpl/consent_manager.py	Live
PDPL Data Rights	services/pdpl/data_rights.py	Live
Audit Service	services/audit_service.py	Live
Audit Log Model	models/audit_log.py	Live
Outbound Governance	services/outbound_governance.py	Live
Tool Verification	services/tool_verification.py	Live
Tool Receipts	services/tool_receipts.py	Live
SLA Escalation Alerts	services/sla_escalation_alerts.py	Live
Skill Governance	services/skill_governance.py	Live

Target State

Component	Status
Contradiction Engine	Building
Saudi Compliance Matrix (live controls)	Building
OPA policy engine	Watch
OpenFGA authorization graph	Watch
Vault secrets governance	Watch

---

4. Data Plane

Purpose: Where data lives, moves, and is enriched.

Current State

Component	Status
PostgreSQL 16 + asyncpg	Live
pgvector embeddings	Live
Redis 7 (cache + broker)	Live
Multi-tenant data isolation	Live
Alembic migrations	Live
Knowledge Service (RAG)	Live
Domain Events	Live
Integration Sync State	Live
30+ SQLAlchemy models	Live
Mem0 memory engine	Live

Data Governance Rules

1. All tables include tenant_id (via TenantModel base)
2. Money fields use Numeric(12,2), never Float
3. Timezone is Asia/Riyadh (UTC+3)
4. Currency defaults to SAR
5. Soft deletes via deleted_at field
6. PII never stored in logs
7. pgvector kept updated (security patches)
8. No external RAG SaaS — PostgreSQL + pgvector + KnowledgeService only

Target State

Component	Status
CloudEvents for event schema	Planned
AsyncAPI for event documentation	Planned
Data quality automated checks	Planned
Lineage/catalog layer	Watch

---

5. Operating Plane

Purpose: Where the system monitors, improves, and governs itself.

Current State

Component	File	Status
Observability	services/observability.py	Live
Self-Improvement Loop	services/self_improvement.py	Live
Feature Flags	services/feature_flags.py	Live
Go-Live Matrix	services/go_live_matrix.py	Live
Operations Hub	services/operations_hub.py	Live
GitHub Actions CI	.github/workflows/dealix-ci.yml	Live
Claude Commands	.claude/commands/	Live
Claude Hooks	.claude/hooks/	Live

Target State

Component	Status
Architecture Brief preflight	Building
Connector Governance Board	Building
Model Routing Dashboard	Building
OIDC authentication	Planned
Artifact attestations	Planned
Audit log external streaming	Planned
Protected branch rulesets	Planned

---

Plane Interaction Rules

1. Decision → Execution: Decision Plane emits structured directives; Execution Plane processes them as tasks.
2. Execution → Trust: Every execution step checks Trust Plane before performing sensitive actions.
3. Trust → Data: Trust Plane reads audit logs and compliance state from Data Plane.
4. Data → Operating: Operating Plane monitors Data Plane health and triggers alerts.
5. Operating → All: Operating Plane can pause, resume, or rollback any plane component.

No plane bypasses Trust for Class B or C actions. This is enforced at the OpenClaw Gateway level.