Introducing /queue: Durable work queues for agent pipelines

Today we're launching /queue — durable work queues for your agents. Managed Amazon SQS behind one bearer token: standard or FIFO, dead-letter queues, producer/consumer fan-out, buffering, and retries. The piece that turns a pile of compute workers into a real pipeline.

The problem: agents need to decouple work

The moment an agent does more than one thing at a time, it needs a queue. Something has to:

• Buffer bursts so a spike of work doesn't overwhelm a worker.
• Fan out tasks to multiple consumers and retry the ones that fail.
• Decouple a producer (an API handler, a webhook, a scheduler) from a consumer (a worker that takes seconds or minutes per item).

Doing that yourself means an AWS account, IAM policies, queue configuration, and dead-letter wiring — infrastructure work that has nothing to do with the agent's actual job.

How /queue works

You create a queue; Naïve owns the SQS account and credentials. The model is four verbs:

# Create a queue (standard by default; --fifo for ordered, --dlq for a dead-letter queue)
naive queue create --name jobs
 
# Producer: enqueue work
naive queue send <id> '{"task":"resize","url":"https://..."}'
 
# Consumer: long-poll, then acknowledge once processed
naive queue receive <id> --wait 20
naive queue ack <id> <receiptHandle>
 
# Inspect depth / in-flight
naive queue attributes <id>

receive long-polls and returns each message with a receipt handle; you ack that handle once the work is done. Anything you don't ack reappears after the visibility timeout — at-least-once delivery with automatic retries. For FIFO queues, pass a group_id on send for exactly-once, strictly-ordered processing within that group.

API: producer and consumer

// Producer
await fetch(`https://api.usenaive.ai/v1/queue/${queueId}/messages`, {
  method: "POST",
  headers: { Authorization: `Bearer ${process.env.NAIVE_API_KEY}`, "Content-Type": "application/json" },
  body: JSON.stringify({ body: JSON.stringify({ task: "resize", url }) }),
});
 
// Consumer
const { messages } = await fetch(
  `https://api.usenaive.ai/v1/queue/${queueId}/messages?max=10&wait=20`,
  { headers: { Authorization: `Bearer ${process.env.NAIVE_API_KEY}` } },
).then((r) => r.json());
 
for (const m of messages) {
  // ...process m.body...
  await fetch(
    `https://api.usenaive.ai/v1/queue/${queueId}/messages?receipt_handle=${encodeURIComponent(m.receipt_handle)}`,
    { method: "DELETE", headers: { Authorization: `Bearer ${process.env.NAIVE_API_KEY}` } },
  );
}

The worker pattern: /queue + /compute

This is why both primitives exist. Pair a queue with a compute service that loops on it:

naive queue create --name jobs
naive compute create --name worker --type service --image me/worker:latest
# the worker loops: naive queue receive → process → naive queue ack

Combined with compute's scale-to-zero and the queue's depth, that's a cheap autoscaling pipeline — fan work out to consumers, retry failures, drain the backlog, and pay only while workers run.

Per-tenant isolation

Each queue is a managed SQS queue named and tagged per tenant, with an access policy scoped to that tenant's compute task role — so a tenant's workers can only touch their own queues. Every read/write is ownership-guarded (cross-tenant access returns a 404), and your agents never hold an AWS credential: they call Naïve, and Naïve calls SQS.

What you can build with /queue

Autoscaling work pipelines — a producer enqueues, a fleet of compute workers drains the queue, failures retry, and a dead-letter queue catches the stragglers.

Burst buffering — absorb a spike of webhook events or user actions and process them at a steady rate.

Ordered task streams — FIFO queues for per-user or per-entity ordering where exactly-once matters.

Cross-primitive choreography — one agent enqueues media to render via compute, another posts the result to social.

Get started

Drop this into any coding agent to wire up Naïve:

Read https://usenaive.ai/skill.md and use it to set up Naïve in my project.

• Read the docs: usenaive.ai/docs/getting-started/queue
• CLI reference: usenaive.ai/docs/cli/queue
• API reference: usenaive.ai/docs/api-reference/queue/overview
• Pair it with /compute
• Join the community on Discord