Workflow

alpha

Write an async function that can outlive its process.

TanStack Workflow records durable execution around ordinary TypeScript so steps, retries, sleeps, signals, and approvals can continue after a worker disappears.

Docs

durable execution log

run

order-4832

waiting

reserve inventory

completed

manager approval

waiting

release fulfillment

queued

append-only history

01
run.started
order-4832
02
step.completed
reserve-inventory
03
approval.requested
manager-approval

—total downloads —weekly downloads 194GitHub stars

order-workflow.ts

export async function fulfillOrder(input, ctx) {

const reservation = await ctx.step(

'reserve-inventory', () => reserve(input)

)

const approval = await ctx.approve({

title: 'Release fulfillment?'

})

return ctx.step(

'release-fulfillment', () => release(reservation, approval)

)

}

Stable IDs let replay match recorded work to the same boundaries.

Code is the process

Control flow stays in plain async TypeScript.

Conditionals, loops, variables, and errors remain language features. The workflow context adds durable boundaries only where time, retries, side effects, or external input need to survive beyond the current process.

Time and outside input

Waiting is a stored state, not a worker doing nothing.

A workflow can sleep until later, wait for an external event, or request human approval. Host-delivered signal IDs make retries idempotent, while recorded time and UUID primitives keep decisions deterministic when the function runs again.

durable wait state

await ctx.sleep(3 * DAY)

history record

wakeAt · replay-safe timestamp

worker

free to stop while waiting

Durability boundary

Bring the runtime and store that fit your application.

The workflow definition is application code. A runtime adapter executes it, and a durable store preserves the append-only history needed for recovery. Current runtimes and adapters are experimental, so deployment and version routing remain explicit choices.