Cron that just runs.
No infra to babysit.
Run any container on a schedule — backups, reports, cleanups, syncs — created and managed by API or your AI agent. The simple, flat-rate alternative to AWS EventBridge, Lambda schedules and hand-rolled crontabs.
A scheduled task on AWS is never just a scheduled task
On AWS a “cron” means EventBridge rules, a Lambda (or a Fargate task), an execution role, CloudWatch Logs and a pricing model with several dimensions — and cold starts, 15-minute limits and IAM in the way. A crontab on a VM is simpler but has no isolation, no logs, no retries and no HA. koigrid gives you the middle: run a real container on a cron, with limits, logs, retries and isolation, in one API call.
How koigrid compares
| koigrid | AWS EventBridge + Lambda | Crontab on a VM | |
|---|---|---|---|
| Run any container/image | Yes | Lambda runtimes / Fargate | Yes (no isolation) |
| Time limit | You set it | 15 min (Lambda) | None |
| Logs + exit code captured | Built-in | CloudWatch (extra) | You wire it up |
| Retries with backoff | Built-in | Configurable | None |
| Isolation | gVisor sandbox | Managed | None (shared host) |
| Pricing | Flat rate | Requests + GB-s + logs | VM cost |
| Create/manage by API or AI agent | Native | Complex (IAM, ARNs) | SSH + edit files |
From public docs, 2026. koigrid jobs run docker run on a shared runner pool with RAM/CPU limits, timeout and captured logs.
Everything a real scheduler needs
Any container, any schedule
Standard 5-field cron with timezone/DST. Run your own image or a shell command — backups, ETL, reports, cleanups.
Retries with backoff
A run that fails is retried with exponential backoff up to your limit — no lost jobs.
Logs, limits and timeouts
stdout/stderr and exit code captured per run; RAM/CPU caps and a kill-timeout so a job never runs away.
gVisor isolation
Each run executes in a syscall sandbox — untrusted code can’t escape to the host.
Agent-native
Create, trigger and inspect jobs from Claude Code or any LLM via the API — full parity with the dashboard.
Anti-overlap
If the previous run is still going, the next occurrence is skipped — no pile-ups.