{"slug":"architecture/12_AGENTS","title":"12 · Agents","section":"architecture","sectionLabel":"Architecture","status":"unclassified","type":"document","path":"architecture/12_AGENTS.mdx","bytes":4040,"description":"The complete agent taxonomy — Domain, Specialist, and System agents that extend Twin capabilities","updated":null,"source":"integratewise/Integratewise-docs-wiki","content":"\n\n**The agent fleet that extends Twin capabilities.**\n\nAgents are specialized AI participants that operate within governance. They are not autonomous actors — they execute under the authority of the Twin and the Governance layer.\n\n---\n\n## Agent Taxonomy\n\n```text\n                    AGENTS\n                       │\n         ┌─────────────┼─────────────┐\n         │             │             │\n    DOMAIN AGENTS  SPECIALIST    SYSTEM AGENTS\n         │         AGENTS            │\n         │             │             │\n    Sales Agent   Research Agent  Orchestration\n    Finance Agent Analysis Agent  Scheduling\n    Support Agent Drafting Agent  Monitoring\n    Account Agent Integration     Reconciliation\n    Operations    Agent           Memory\n```\n\n---\n\n## Domain Agents\n\n| Agent | Domain | Capabilities |\n|-------|--------|-------------|\n| **Sales Agent** | Sales | Pipeline analysis, deal scoring, next-best-action |\n| **Finance Agent** | Finance | Revenue analysis, billing review, collection priority |\n| **Support Agent** | Support | Ticket triage, SLA prediction, escalation routing |\n| **Account Agent** | Account Success | Health analysis, renewal risk, expansion opportunity |\n| **Operations Agent** | Operations | Workflow optimization, resource allocation |\n\n### Domain Agent Rules\n\n- Scoped to their business domain\n- Reason and recommend within their domain\n- Escalate cross-domain work to the orchestrator\n- Feed learnings back to the Spine\n\n---\n\n## Specialist Agents\n\n| Agent | Specialty | When Used |\n|-------|-----------|-----------|\n| **Research Agent** | Deep investigation | When context is insufficient |\n| **Analysis Agent** | Quantitative analysis | When numbers need interpretation |\n| **Drafting Agent** | Content creation | When output needs to be produced |\n| **Integration Agent** | System integration | When external systems need interaction |\n\n### Specialist Agent Rules\n\n- Invoked on demand\n- Receive focused context for their task\n- Return results to the calling agent\n- Do not maintain persistent state\n\n---\n\n## System Agents\n\n| Agent | Function |\n|-------|----------|\n| **Orchestration Agent** | Coordinates multi-step workflows |\n| **Scheduling Agent** | Manages time-based triggers |\n| **Monitoring Agent** | Observes system health |\n| **Reconciliation Agent** | Verifies execution outcomes |\n| **Memory Agent** | Manages organizational memory |\n\n### System Agent Rules\n\n- Operate at the platform level\n- Not directly visible to users\n- Enforce governance and policies\n- Fully auditable\n\n---\n\n## Agent Communication (ACP)\n\nAgents communicate via the Agent Communication Protocol:\n\n```typescript\ninterface ACPMessage {\n  from: AgentRef;\n  to: AgentRef;\n  workItem: WorkItemRef;\n  context: ContextSnapshot;\n  intent: string;\n  constraints: Constraint[];\n  deadline?: string;\n}\n```\n\n### ACP Rules\n\n- Hand-offs are logged in the Spine\n- Context is transferred, not lost\n- Receiving agent can accept, reject, or delegate\n- High-risk transfers require governance approval\n\n---\n\n## Agent Security Boundary\n\n```text\nCanonical Spine\n      ↓\nProjection (role + authorization)\n      ↓\nContext Assembly (relevance + token budget)\n      ↓\nAgent Context (sandbox)\n      ↓\nAgent Execution\n```\n\nAgents operate against authorized projections, never unrestricted canonical Spine.\n\n---\n\n## Agents Connect to\n\n| Layer | Relationship |\n|-------|-------------|\n| [Twin Workbench](/architecture/07_TWIN_WORKBENCH) | Twin delegates to agents |\n| [Work Allocation](/architecture/09_WORK_ALLOCATION) | Work assigned to agents |\n| [Governance](/architecture/10_GOVERNANCE) | Agent actions gated by governance |\n| [Communication Bridge](/architecture/11_COMMUNICATION_BRIDGE) | Agent communication via ACP |\n| [Execution](/architecture/13_EXECUTION) | Agents execute via Execution Fabric |\n"}