@reactive-agents/orchestration

Multi-agent orchestration for the Reactive Agents framework. v0.10.3

Coordinate fleets of agents through workflow patterns: sequential pipelines, parallel fan-out / fan-in, map-reduce, and durable event-sourced executions. The WorkflowEngine runs workflows as Effects with typed message passing, checkpoint persistence, and per-step failure isolation.

Installation

bun add @reactive-agents/orchestration

Or via the umbrella:

bun add reactive-agents

Features

• Workflow patterns — WorkflowPattern enum exposes sequential, parallel, pipeline, map-reduce, agent-as-tool
• Worker pool — pool of registered WorkerAgents, dispatched per-step
• Durable execution — EventSourcing records every state transition; resume from checkpoints
• Typed handoffs — Effect Schemas validate inter-agent messages
• Failure isolation — one agent failing emits WorkflowStepError; the workflow can continue, retry, or abort per the policy
• A2A integration — pairs with @reactive-agents/a2a for cross-process agent communication

Quick Example

import { Effect } from "effect";
import {
  OrchestrationService,
  OrchestrationServiceLive,
  WorkflowPattern,
} from "@reactive-agents/orchestration";

const program = Effect.gen(function* () {
  const orch = yield* OrchestrationService;

  const workflow = yield* orch.createWorkflow({
    name: "research-pipeline",
    pattern: WorkflowPattern.Sequential,
    steps: [
      { id: "search",     agentId: "searcher",   input: (ctx) => ctx.query },
      { id: "summarize",  agentId: "summarizer", input: (ctx) => ctx.search },
      { id: "critique",   agentId: "critic",     input: (ctx) => ctx.summarize },
    ],
  });

  return yield* orch.runWorkflow(workflow.id, {
    query: "Latest advances in fusion energy",
  });
});

await Effect.runPromise(program.pipe(Effect.provide(OrchestrationServiceLive)));

Patterns

| Pattern | Behaviour | | -------------- | -------------------------------------------------------------------- | | sequential | Step N+1 receives the output of step N | | parallel | All steps run concurrently against the same input | | pipeline | Streaming hand-off — outputs flow as soon as available | | map-reduce | Fan-out per input element, then reduce results | | agent-as-tool | One agent invokes another as a tool inside its own kernel loop |

Durable Execution

import { makeEventSourcing } from "@reactive-agents/orchestration";

const sourcing = yield* makeEventSourcing({ store: mySqliteStore });
// every workflow state transition is appended; resume by replaying events
const resumed = yield* sourcing.replay(workflowId);

Key Exports

| Export | Purpose | | ------------------------------------------------- | -------------------------------------------------- | | OrchestrationService, OrchestrationServiceLive | Composite orchestration entry point | | makeWorkflowEngine | Pluggable workflow engine | | makeEventSourcing | Durable event log for workflow state | | makeWorkerPool | Worker-agent registry + dispatcher | | createOrchestrationLayer | Factory for the runtime layer | | WorkflowPattern, WorkflowState, Workflow, WorkflowStep, Checkpoint, WorkerAgent | Schemas + types | | WorkflowError, WorkflowStepError, CheckpointError, WorkerPoolError | Tagged errors |

Documentation

• Full docs: docs.reactiveagents.dev
• Orchestration guide: docs.reactiveagents.dev/guides/orchestration/
• Pairs with @reactive-agents/a2a for cross-process agent comms

License

MIT