Architecture Overview

ODIN uses a two-layer framework designed for extensibility and clear separation of concerns.

Two-Layer Framework

┌─────────────────────────────────────────────────────────────────┐
│ LAYER 1: THE AGENTS (the constant)                              │
│                                                                 │
│  ┌─── Front Office ───┐  ┌─ Middle Office ─┐  ┌─ Back Office ──┐│
│  │ Intelligence       │  │ Risk mgmt  [IND]│  │ Operations     ││
│  │ Origination   [NET]│  │ Compliance [IND]│  │ Trade finance  ││
│  │ Merchant      [NET]│  └─────────────────┘  │ Settlement     ││
│  │ Hedging            │                       └────────────────┘│
│  └────────────────────┘                                         │
│                                                                 │
│  [NET] = network-enabled (ODIN trade discovery)                 │
│  [IND] = structurally independent (can override commercial)     │
├─────────────────────────────────────────────────────────────────┤
│ LAYER 2: THE SERVICES (the variables, all MCP-gated)            │
│                                                                 │
│  Trade discovery (ODIN network) | Counterparty data             │
│  Compliance screening | Market pricing | Logistics & tracking   │
│  Document services | FX & payments | Weather & climate          │
│  Financial services (loan origination)                          │
│                                                                 │
│  Each: free tier (rate-limited) → paid tier (full access)       │
│  New service = new MCP endpoint. Architecture never changes.    │
└─────────────────────────────────────────────────────────────────┘

Agent Topologies

Sub-agent Topology: Star (Human at Center)

Every connection goes through the human. The human invokes each agent one at a time, reviews output, then invokes the next. No agent sees what another agent produced unless explicitly shared.

Team Agent Topology: Mesh (Orchestrator Coordinates)

Agents can communicate directly. Risk and Compliance have override authority and can block commercial decisions. Settlement data feeds back to Risk (credit scores), Intelligence (outcome data), and Operations (supplier scoring).

Seven-Phase Trade Lifecycle

Phase	Duration	Key Activities
1. Origination	Hours to months	Discovery, opportunity evaluation
2. Evaluation	Hours to 4 weeks	Risk/compliance screening, parallel assessments
3. Execution	Same day to 2 weeks	Contract documentation, hedge placement
4. Logistics	1-8 weeks	Freight, insurance, inspections
5. Delivery	1 day to 6 weeks	Loading, transit, document assembly
6. Settlement	5-90 days	Invoicing, payment, reconciliation
7. Closeout	Weeks to YEARS	PPA monitoring, audit trail

Override and Escalation

Twelve override triggers implemented in team agent mode:

VaR soft breach (>80%) - Alert
VaR hard breach (100%+) - Mandatory reduction
Stop-loss breach - Immediate hedging
Credit limit breach - Block execution
Credit quality / concentration - Hold + mitigation
Liquidity / margin stress - Block new trades
Sanctions / KYC incomplete - Compliance hold
Reputational concern - Route to committee
Investigation / whistleblowing - Freeze + preserve trail
Stress test failure - CRO notification
Rogue trading suspicion - Trading suspended
Economic capital breach - Exception required

Data Continuity

Four capture mechanisms ensure the agents maintain context:

Tool Integrations (Gmail, Calendar, Drive via MCP)
Agent-prompted updates (follow-up after milestones)
Voice-first quick capture (voice debriefs)
Document-as-trigger (BOL, quality certs, etc.)