Hello muhammad_iqbal;

yes it's possible to self-host vercel workflows on railway without using vercel at all

the workflow sdk is open source and has an official package called @workflow/world-postgres made exactly for this , self-hosting on platforms like railway, render, fly.io etc. it uses postgres for state storage and graphile-worker for job processing

here's what you need:

first, install the packages: npm install workflow @workflow/world-postgres

then, set these two env vars in railway (railway already gives you DATABASE_URL from its postgres plugin which the sdk falls back to automatically): WORKFLOW_TARGET_WORLD="@workflow/world-postgres" and WORKFLOW_POSTGRES_URL="your railway postgres url"

after that, run the db migration once: npx workflow-postgres-setup

at the end, in your next.js project create an instrumentation.ts at the root with this:

export async function register() { if (process.env.NEXT_RUNTIME !== "edge") { const { getWorld } = await import("workflow/runtime"); const world = await getWorld(); await world.start?.(); } }

that's it railway works perfectly for this because it runs long-running servers by default, which is exactly what this needs (it won't work on serverless)

Hope this help you :)

my info source is from here ; workflow-sdk.dev/worlds/postgres and railway is explicitly listed as a supported deployment target on that page

Thanks!

To scale up, I think there should be anyway we can add more workers or deploy workflows as separate process. Is there any guide/way for that?

there are two confirmed ways to scale:

option 1 is to increase concurrency within the same process via env var : set WORKFLOW_POSTGRES_WORKER_CONCURRENCY to whatever number you want (default is 10). the docs also say to set WORKFLOW_POSTGRES_MAX_POOL_SIZE to queueConcurrency + 2 when going higher

option 2 is a programmatic config if you want more control : in workflow.config.ts you can do createWorld({ connectionString: "...", queueConcurrency: 20, maxPoolSize: 22 })

option 3 is horizontal scaling, so since everything is backed by postgres + graphile-worker, you can just run multiple instances of your app (multiple railway services) all pointing to the same postgres database. graphile-worker is designed to handle multiple workers polling the same queue safely, so this works out of the box on railway by just scaling your service replicas

source is the same page: workflow-sdk.dev/worlds/postgres, configuration section

Hope this help you :)