Durable workflows,
that survive anything.
Define a workflow as ordered steps. Each step’s result is journaled, so a run survives crashes and redeploys without re-running completed steps — plus durable sleep (seconds to months) and retries. The agent-native, no-lock-in alternative to AWS Step Functions.
Long-running work shouldn’t break on a redeploy
Multi-step jobs — onboarding flows, order pipelines, agent tasks — need to survive crashes, redeploys and long waits. Step Functions solves this but means Amazon States Language, the console and per-transition pricing. koigrid Workflows give you the core: define ordered steps, start a run, and your worker asks for the next step, runs it and reports the result. Each result is journaled, so a crashed or redeployed worker resumes exactly where it left off — completed steps never re-run. Durable sleep pauses a run for seconds to 90 days; failed steps retry with backoff. One open API your agent can drive end to end.
How koigrid compares
| koigrid | AWS Step Functions | |
|---|---|---|
| Definition | Ordered steps (API/CLI) | Amazon States Language (JSON) |
| Survives crash/redeploy | Yes (journaled steps) | Yes (managed state) |
| Durable sleep | Seconds to 90 days | Wait state |
| Retries + backoff | Built in (per step) | Built in |
| Pricing | Flat (included) | Per state transition |
| Lock-in | None (open API + Postgres) | ASL + AWS-specific |
From public docs, 2026. Journaled-step model (like Step Functions activities / Inngest steps), not deterministic-code replay.
What you’d pay elsewhere
1M workflow runs/month with retries and waits
koigrid Workflows (Starter)
€12/mo/mo
flat — workflows + runs included in your plan
AWS Step Functions
~$25+/mo/mo
per-state-transition pricing that grows with steps × runs
flat & predictable
Prices are indicative; koigrid is flat-rate and EU-hosted.
Everything you need for durable, long-running work
Journaled steps
Each step’s output is persisted — a run resumes after a crash or redeploy without re-running completed steps.
Durable sleep
Pause a run for seconds to 90 days; it wakes up and continues. Survives restarts (state lives in Postgres).
Retries with backoff
A failing step retries with exponential backoff up to maxAttempts, then the run fails cleanly.
At-least-once steps
A claimed step has a lease — if the worker dies mid-step, the lease expires and another worker picks it up.
Any worker
Your worker (or agent) asks for the next step, runs it however it wants, and reports complete/fail. No SDK lock-in.
Agent-native
Start and drive a durable, resumable workflow from Claude Code or any LLM via the API + a scoped token.