Orchestration & governance

Coordinate agents and workflows with control

Where agent orchestration meets enterprise integration: long-running tasks, events, compensation, policy enforcement and auditability.

🧠

Multi-agent orchestration New

Supervisor, hand-off, fan-out, event-driven and workflow-led patterns, with a worked invoice example.

Read guide →
◈

Integration orchestration

Dependencies, retries, compensation, correlation IDs and event processing.

Read guide →
⏱

Long-running agent tasks

A2A task states, push notifications and MCP's tasks extension, mapped to asynchronous integration patterns.

Read guide →
✋

Human-in-the-loop

Approvals as separate, logged requests using MCP Multi Round-Trip Requests and A2A input-required.

Read guide →
🛡

Governance & security policies

Policy types, enforcement points, human approval, agent lifecycle and audit evidence.

Read guide →
⚠

Agentic risk controls

Cascading failures, rogue agents and tool misuse, controlled in gateways and iFlows.

Read guide →
📊

Reliability & observability

OpenTelemetry trace context through agents, gateways and integration flows.

Read guide →

Patterns that carry over from classic integration

Agentic situationIntegration pattern
Several agents must finish one business processOrchestrator or workflow-led backbone with A2A delegation, depth limits and budgets
Agent retries a write after a dropped streamIdempotent receiver with idempotency keys
Step 3 of 5 fails after an agent posted documentsSaga with compensating actions
A2A task runs for hoursAsynchronous request-reply with correlation ID and status endpoint
Agent floods a backend with callsThrottling, circuit breaker, bulkhead
Business event should wake an agentEvent-driven consumer via an event broker, not polling
High-impact action needs approvalHuman task / approval step with timeout and escalation