AIF-Orchestrator

Multi-Agent Coordination Engine — DAG workflow execution, retry/rollback, human-in-the-loop (HITL) approval gates, and Saga compensation patterns.

Capabilities

• DAG Execution: Compose agent tasks into directed acyclic graphs with dependency resolution
• Saga Compensation: Distributed transaction patterns with automatic rollback on failure
• HITL Gates: Policy-driven approval checkpoints requiring human sign-off
• Retry Engine: Exponential backoff with jitter and max attempts config
• Checkpoint/Restore: Save workflow state; resume from last checkpoint on failure
• Parallel Dispatch: Execute independent tasks concurrently across the agent mesh

Workflow Pattern

Planner → [Task A, Task B] → Verifier → [Task C] → Executor
                ↓ (parallel)                        ↓
            Skill Registry                     Receipt Store

Governance Integration

Every orchestration step passes through the 6-gate pipeline:

1. Intent parsing
2. Risk classification (LOW / MEDIUM / HIGH / CRITICAL)
3. Policy matching
4. Approval routing
5. Capability dispatch
6. Immutable receipt

Example

from runtime.workflow_orchestrator import WorkflowOrchestrator

orchestrator = WorkflowOrchestrator()
workflow = orchestrator.create(
    tasks=["research_competitor", "analyze_market", "generate_report"],
    dependencies={"generate_report": ["research_competitor", "analyze_market"]}
)
result = orchestrator.execute(workflow)

Repository

https://github.com/frostyjay7813/AIF