Orchestration Adapters

Adapters decide where orchestration runs execute and how durable they are. Task and workflow definitions do not change when you switch adapters.

Memory

Use createMemoryAdapter() when you want:

• no external infrastructure
• fast local development
• lightweight tests

Tradeoffs:

• no durability across process restarts
• no cross-process coordination

const adapter = createMemoryAdapter({ concurrency: 10 });

SQLite

Use createSqliteAdapter() when you want:

• durable local or single-node execution
• restart recovery without Redis
• persistent run and step state

Tradeoffs:

• still a single-process or single-node shape
• not the right fit for distributed worker fleets

const adapter = createSqliteAdapter({
  path: './data/orchestration.db',
  concurrency: 10,
});

BullMQ

Use createBullMQOrchestrationAdapter() when you want:

• Redis-backed durability
• repeatable schedules
• BullMQ workers and queues
• progress updates through QueueEvents

Tradeoffs:

• requires Redis
• signals are not supported
• you now operate queue infrastructure

import { createBullMQOrchestrationAdapter } from '@lastshotlabs/slingshot-orchestration-bullmq';

const adapter = createBullMQOrchestrationAdapter({
  connection: { host: '127.0.0.1', port: 6379 },
  prefix: 'orch',
  concurrency: 10,
});

When used through createOrchestrationPackage(), the plugin starts and shuts down the adapter for you. When used directly through createOrchestrationRuntime(), the adapter now lazily starts on first use, so runTask() / runWorkflow() do not require a separate manual start() call.

Temporal

Use createTemporalOrchestrationAdapter() from @lastshotlabs/slingshot-orchestration-temporal when you want:

• distributed workers across multiple machines
• durable workflow history, timers, retries, and signals
• Temporal-native scheduling and visibility
• Slingshot task/workflow definitions running on a real Temporal cluster

Tradeoffs:

• requires an existing Temporal service or Temporal Cloud namespace
• requires a separate Node-based worker process
• operationally heavier than SQLite or BullMQ

import { Client, Connection } from '@temporalio/client';
import { createTemporalOrchestrationAdapter } from '@lastshotlabs/slingshot-orchestration-temporal';

const connection = await Connection.connect({ address: 'localhost:7233' });
const client = new Client({ connection, namespace: 'default' });

const adapter = createTemporalOrchestrationAdapter({
  client,
  connection,
  namespace: 'default',
  workflowTaskQueue: 'slingshot-workflows',
  defaultActivityTaskQueue: 'slingshot-activities',
  ownsConnection: false,
});

Temporal worker processes are separate from the HTTP server. For real deployments, pair the adapter with createTemporalOrchestrationWorker() so workflow code and activity handlers can poll the configured task queues.

Capability matrix

Capability	Memory	SQLite	BullMQ	Temporal
Core execution	Yes	Yes	Yes	Yes
Scheduling	No	No	Yes	Yes
Run listing	Yes	Yes	Yes	Yes
Progress subscriptions	Yes	Yes	Yes	Yes
Signals	No	No	No	Yes

Not available yet

These providers are planned but not shipped in the current worktree:

• Trigger.dev
• Step Functions

If you need those today, the answer is still “not implemented”, not “hidden behind a flag”.