@agentium/queue - Agentium

Run agents, teams, and workflows in Redis-backed background workers. Start with the queue quickstart to run a complete example.

npm install @agentium/core@4.0.0 @agentium/queue@4.0.0 bullmq@^6.3.11 ioredis

BullMQ and ioredis are optional peers: install them when you use this adapter. BullMQ ^5.81.5 is also supported. Existing legacy repeat schedules need the migration procedure before moving to v6.

Find a signature

This reference covers all 18 named exports of @agentium/queue at 4.0.0, including public methods, overloads, and configuration fields.

API section	Exports	Learn to use it
Producer and schedules	3	Guide
Workers and progress	3	Guide
Jobs and status	6	Guide
Durable delivery	6	Guide

Runtime behavior

Concern	Contract in v4
Queue identity	Producer and worker must share the Redis connection and `queueName`; the default name is `agentium-jobs`.
Startup	`new AgentWorker(...)` begins consuming. `start()` is a compatibility no-op. Default concurrency is 5.
Retries	Set `attempts` and `backoff` on enqueue options or producer `defaultJobOptions`. A retry runs the job again and may repeat external effects.
Progress	Agent progress counts `run.stream.chunk` events for that run. It is not a completion percentage.
Cancellation	`cancelJob()` calls BullMQ job removal. It does not signal a running agent and can reject for a locked active job.
Missing jobs	`getJobStatus()` throws when a job is absent, including after removal.
Shutdown	`await worker.stop()` drains active work, with a default 30-second timeout. `await queue.close()` closes producer connections.

Ordinary jobs call Agent.run(), Team.run(), or Workflow.run(). Durable delivery uses task envelopes, leases, and registered replay-aware drivers. Choose it when you need persisted task recovery, and design side effects for replay.