open-source job queue · AI layer on top

Your failed jobs come back with answers, not stack traces.

Relay runs your background jobs for Node.js — Postgres and Redis underneath. When a job fails, the AI layer returns the root cause in plain English, a confidence score, and a suggested fix. It even predicts overload before it happens.

View on GitHub
  • Node.js + TypeScript
  • PostgreSQL + Prisma
  • Redis + BullMQ
  • REST API
relay — dashboardCreate Job

Total Jobs

14

Success Rate

86%

Active Workers

5 / 5

Real-time activity

completedprocessingfailed

AI Insights

86%

Health score

No active predictions right now.

Recent jobs

  • send-welcome-emailcompleted
  • generate-invoiceactive
  • process-paymentcompleted
  • sync-crm-contactsfailed
  • resize-avatarcompleted
explained in plain english◆overload predicted before it happens◆chain queues into workflows◆ai actions need your approval◆create a queue◆post jobs over rest◆workers process◆transient failures retry themselves◆

Every queue has silent failures. Relay refuses to stay silent about them.

the pain

A job fails at 3 a.m. All you get is a 200-line stack trace.

Root cause diagnosis

The AI layer reads the failure and returns the root cause in plain English, a confidence score, and a suggested fix — right in the dashboard.

the pain

Retries hammer a broken downstream API all night because the error isn't transient.

Intelligent retries

Transient failures retry automatically. Non-transient ones get flagged with the suggested fix instead of blindly retrying.

the pain

A queue quietly backs up. Nobody notices until it's an incident.

Predictive monitoring

Relay predicts capacity problems — a queue heading toward overload — and warns you before it becomes one.

how it works

From create a queue to explained failure in three steps.

01

Create a queue

One REST call (or a click in the dashboard) stands up a queue on Postgres and Redis.

POST /v1/queues
{ "name": "emails" }
02

Post jobs — workers process them

Push jobs over the REST API or the in-repo TypeScript SDK. Your workers pick them up; transient failures retry on their own.

POST /v1/jobs
{ "queueId": "q_emails",
  "name": "send-welcome",
  "payload": { "userId": 402 } }
→ 201 · queued
03

See the explanation, not the stack trace

When a job fails, the AI analysis appears in the dashboard: root cause, confidence, suggested fix.

// root cause (96% confidence)
SMTP auth failed — the password reset on Mar 12 rotated your credentials. Non-transient; retried 3x.
fix · update SMTP password, then re-run the job

features

A queue that doesn't just run your jobs — it understands them.

the standout

Operate your queues by chatting

Ask about your jobs in plain English. The agent answers — and it can act: pause a queue, cancel a job. Anything destructive waits for your explicit approval.

Why is my emails queue backing up?

The emails queue is at 82% capacity — the pdf-render step upstream slowed down 4x in the last hour, so jobs are arriving faster than workers drain them. Add a worker or raise concurrency for emails.

Why is my queue backing up?Pause the emails queue

Failures, explained

The AI error-analysis engine reads the failure and returns a root cause, a confidence score, and a suggested fix.

root cause · 96% confidence

The job reads user.id but the user record doesn't exist — accounts deleted in the last hour still have queued jobs.

fix · check the user exists before enqueueing, or prune queued jobs on account deletion

non-transient · retries stopped automatically

Multi-step workflows

Chain queues into a sequence. Relay advances each step automatically — and halts the run the moment a step fails.

1charge-card
2generate-invoice
3send-email

one run · three queues · relay advances each step

capacity predictions ride along too — a queue heading toward overload gets flagged early.

Every workspace is isolated

Every signup gets its own workspace: its own queues, its own API keys, its own usage limits. Full multi-tenant separation, zero bleed.

your workspace

Queuesyours alone
API Keysrelay_sk_live_••••
Usage limitsper-workspace

open source · no lock-in

Built by one engineer, in the open.

Relay doesn't have a customer logo wall — it has a public repository. It's a solo-built, early-stage project: every commit, every issue, and the whole roadmap are on GitHub. If it disappears tomorrow, you still have the code.

Read the code

One public repo

The backend, the dashboard and the TypeScript SDK all live on GitHub.

No lock-in

A REST API on top of Postgres — self-host it, or migrate off it.

Free during beta

There's no billing system yet, so there's no pricing page.

Honest by default

No logo wall, no testimonials, no made-up uptime numbers.

reviews · none yet

No reviews yet — we just launched.

Relay doesn't have customer testimonials because it doesn't have customers yet. This section will fill up with real feedback from real users. If you try it, we'd genuinely like to hear what broke.

Fair questions, straight answers.

Is Relay free to use?+

Yes — there's no billing system built yet, so the hosted beta is free. Create a workspace and use it. When paid plans eventually exist, this page will say so honestly.

Can I self-host it?+

Yes. Relay is a Node.js + TypeScript backend and a Next.js dashboard on top of PostgreSQL and Redis, all in one public repo — clone it, start Postgres and Redis (a docker-compose file is included), and run both apps. Two parts are external services: dashboard sign-in uses Clerk, so you'll need your own Clerk keys, and the AI features call OpenAI — without an OpenAI key they simply switch off.

Do I have to use the AI features?+

No. The AI layer runs alongside your jobs, never in their path. If it's switched off in Settings, unavailable, or your workspace hits its monthly AI limit, jobs still run, retry and complete exactly as they would otherwise — they just don't come back with an explanation.

What happens to my data?+

It stays in your own workspace. Your queues, jobs, logs and API keys belong to that workspace, and every API request is scoped to it, so no other workspace can read them. When AI features are on, a job's name, payload and error message are sent to OpenAI to classify and analyze it — turn those features off in Settings if that isn't acceptable for your data.

Can the AI chat agent delete my stuff?+

No. The agent has no tool for deleting anything, and its destructive actions — cancelling a job or pausing a queue — always stop and wait for your explicit approval before they run.

Your next failed job could come back explained.

Create an isolated workspace in seconds — or clone the repo and run it yourself.

View on GitHub

free during beta · no billing system, so no pricing table · your workspace, your data