- The question that decides everything: where does the state live?
- Four forks that narrow it down
- Scripts, flows and the database route
- What changed in this category
- What durable workflow tools cost
- Kill a worker and watch what happens
- Mistakes that bite after the switch
- Nine tools to look at instead of Inngest
- Which type to choose
Durable workflow tools differ most on where workflow state lives and who runs the engine, so a like-for-like swap is rare. Sort the alternatives by approach first: a durable execution platform, a TypeScript task runner, a library that keeps state in your own Postgres, or workflows tied to the serverless platform you already use.
The best Inngest alternative is whichever tool keeps your workflow state in the place you're happiest running it. For long, failure-proof workflows that's a durable execution platform; for TypeScript background jobs and agents, a task runner; for apps already on Cloudflare or a serverless host, that platform's own workflows; and if you'd rather not run another service at all, a library that keeps the state in your own Postgres.
That framing matters more than any feature grid, because nearly every tool here retries, schedules and shows you a run history. Where they really differ is in who holds the record of a half-finished job, and what happens to it when something dies at 3am.
The question that decides everything: where does the state live?
Inngest runs your code as steps inside functions, triggered by events, and keeps the progress of each run on its side. When a step fails, it retries that step; when a worker restarts, the run carries on. Every alternative gives you some version of that promise, but they keep the record in very different places.
A durable execution platform keeps it in its own engine, which you either run yourself or rent as a cloud service. A task runner keeps it in its own queue and run store. Platform-native workflows keep it inside the hosting platform, so it's close to your code and invisible to you. And the database route flips the model: DBOS says it turns your existing Postgres into the orchestration server, so state sits in tables you already back up, query and understand.
Neither model is wrong. They fail differently, and that's what you're choosing between.
| Approach | Where workflow state lives | Who runs the engine | On this list |
|---|---|---|---|
| Durable execution platforms | The platform's own engine or its cloud | You, or the vendor's hosted service | Temporal, Restate, Orkes Conductor |
| Task runners and queues | The runner's queue and run store | The vendor, or you if self-hosted | Trigger.dev, Hatchet |
| Workflows tied to a platform (Platform-native workflows) | Inside your hosting platform | The platform | Cloudflare Workflows, Upstash Workflow |
| Script platforms | The platform's database | You or the vendor | Windmill |
| The database route | Your own Postgres | Nobody extra; it's a library | DBOS |
Four forks that narrow it down
How long do your workflows run? If a job finishes in a few minutes, almost anything works. If it waits days for a human approval or a webhook, look hardest at the Durable execution platforms, which are built for code that sleeps for a long time and wakes up with its variables intact.
What language is your team writing? Task runners shine in TypeScript codebases: you write a function, add retries and a queue, and ship. Task runners and queues are quick to adopt for background jobs, but check how far the model stretches before you hang a long, branching business process on one.
Where does your app already run? If it's on Cloudflare Workers or a serverless host, the platform's own workflows are the shortest move by a mile, with very little to set up. The trade is lock-in: move hosting later and the workflows have to move too.
Hosted or self-run? Self-host if you need control over where data and workers run and you have someone to keep the engine alive. Otherwise a hosted service takes a whole category of pager alerts off your team. Several tools here are open source with a paid cloud, so you can start one way and switch later.
Scripts, flows and the database route
Two tools on this list don't fit the usual shape. Windmill is a self-hostable home for scripts, flows, schedules and webhooks, closer to an internal automation platform than a code library. It's handy when operations people want a visual view of what ran and when. Script platforms feel less natural when your durable logic sits deep inside an application's own code.
DBOS goes the other way. It's a library, not a service, and it keeps workflow state in Postgres. If your team already runs Postgres well, that removes a whole moving part. If your database is already the bottleneck, think twice before adding workflow traffic to it.
What changed in this category
AI agents changed the pitch. Most of the makers here now describe an agent as one more kind of long-running workflow: a loop of model calls and tool calls that needs retries, timeouts and a record of every step. That's a fair description, and it's why durable execution moved from a back-end niche into the AI conversation. Just don't pick a tool because its home page says agents. The same tests apply whether a step calls a model or charges a card.
What durable workflow tools cost
Nearly all of them let you start without paying. Each platform's own pricing page sets out what its free allowance covers. After that, check what each one counts as a single unit, whether actions, runs or steps, because that decides how the bill grows. Team and enterprise pricing is on the same pages.
Read the unit carefully. One tool's run can be another tool's ten steps, so the same workflow can cost very different amounts. Price your three busiest workflows on each shortlisted tool, not the headline rate. Self-hosting the open source engines costs nothing in licences and quite a lot in someone's attention.
Kill a worker and watch what happens
Spend an hour on this before you sign up for anything. Take one real job with several steps and at least one side effect, sending an email is ideal, and build it in two tools from the approach that fits. Make every step safe to run twice. Start the job, then stop the worker in the middle of a step, the way a deploy or a crash would.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Now restart it and look closely. Did the run resume at the right step? Did the email go out once, or twice? Open each tool's view of the failed run: how quickly could you tell what broke and where? Then try one change that will happen in real life, adding a step to a workflow while an old run is still in flight, and see whether the tool copes or makes you version anything by hand. Whichever one made all of that dull is your pick.
Mistakes that bite after the switch
- Side effects bundled into a step with other work. If a step sends an email and then fails on the next line, the retry sends it again. Give each side effect its own step and make it idempotent.
- Expecting a like-for-like port. Each tool shapes workflows differently, so plan to rewrite the orchestration code and keep business logic in plain functions you can reuse.
- Forgetting scheduled jobs. Cron-style work is easy to miss in a migration and fails silently: it simply stops firing.
- Running both systems for months. Pick a cut-over date for each workflow so no job runs twice from two engines.
- Choosing a self-hosted engine without naming who keeps it alive at night.
Nine tools to look at instead of Inngest
Temporal
A widely used durable execution engine, open source with a hosted cloud, for workflows that must survive crashes and run for days.
Restate
A smaller durable runtime built on a distributed log, aimed at low-latency steps across agents, workflows and microservices.
Orkes Conductor
A commercial platform built on Conductor OSS, pitched at enterprises wanting deterministic, explainable agent and workflow orchestration.
Trigger.dev
An open source TypeScript platform for long background tasks and AI agents, with retries, queues and run observability included.
Hatchet
A task and workflow orchestrator for background jobs and agents that lets you bring your own stack and compute.
Cloudflare Workflows
Durable multi-step execution built into Cloudflare Workers, the obvious first look if your app already lives there.
Upstash Workflow
Turns serverless API routes into durable multi-step processes, saving each finished step so a failure retries only the step that broke.
Windmill
An open source, self-hostable workspace for scripts, flows, schedules and webhooks, a natural fit for operations and internal tools teams.
DBOS
An open source library that keeps workflow state in the Postgres database you already run, so there's no orchestration server to host.
Which type to choose
Long-running, must-not-fail business processes: a durable execution platform, hosted unless you have a team that wants to run it. A TypeScript app with background jobs and agents: a task runner, trialled with the kill-a-worker test above. An app on Cloudflare or a serverless host: that platform's workflows first. A team with strong Postgres skills and no appetite for another service: the database route. Ops-led automation with schedules and webhooks: a script platform.
Pick the approach that matches where your app already runs, then trial two tools inside it. If you're on a serverless platform, the platform-native option is the obvious first test; if you need workflows that run for days and survive anything, look at the durable execution platforms. My Temporal alternatives list goes deeper on that group, and the Airflow alternatives list helps if some of your jobs are really data pipelines.
Do you make a durable workflow or background job tool? Tool makers can ask to be considered for a featured place on this list.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.
