Plans & limits
During the beta
Overnite is in a closed beta, and the beta is free: there is nothing to pay us, no card to add, and no plan to pick. What you pay for is the model usage your agents run up, and that goes to your model provider, on your key or your subscription - never through us, and never marked up.
Before anything costs money we tell you by email at least 30 days ahead, and nothing moves to a paid plan without you choosing it.
Plans after the beta
What overnite.dev/pricing describes today. None of these limits is enforced during the beta, and the details may still change.
| Free | Pro | Team | |
|---|---|---|---|
| Price | $0 | $19 / month | $49 / seat / month |
| Workspaces | 1 | 5 | shared runners |
| Compute | 10 hours / month | 100 hours / month | - |
| Your own model keys and subscriptions | yes | yes | yes |
| Push notifications, scheduled runs | yes | yes | |
| Usage and cost dashboard, audit log | yes |
Compute beyond a plan is planned at $0.10 per hour. A workspace counts compute only while it is awake; a sleeping workspace costs nothing but its disk.
Limits
These apply to everyone today.
Machines
| CPU | 2 cores guaranteed, up to 4 |
| Memory | 4 GB guaranteed, up to 8 GB |
| Workspace disk | 20 GB, kept while the workspace sleeps - not enforced yet, so keep an eye on build caches |
Scratch space (/tmp) | 10 GB, cleared when it sleeps |
Every workspace gets the same machine for now; the size the API accepts is
recorded, not yet applied.
Sleep
| Idle | a workspace sleeps after 30 minutes with every agent idle, suspended or exited, and no output or input |
| Working | a working agent keeps its workspace awake, however long it takes |
| Waiting for you | a blocked agent keeps it awake for up to 6 hours; then the workspace sleeps and the agent is suspended, its question kept for when you resume it |
| Waking | a workspace that has not woken within 10 minutes fails the run that woke it |
Agents are suspended, not killed, when a workspace sleeps: they are given 10 seconds to save their session, and resume where they left off when it wakes (Agents).
Agents and runs
| Agents per workspace | no fixed limit; the machine's CPU and memory are |
| Task or input | 64 KB at a time |
| Readiness | input waits up to 15 seconds for an agent's screen to be recognised, then is sent anyway |
| Schedules | five-field cron, at most every 15 minutes |
| Scheduled or GitHub run that never starts | given up after 30 minutes |
| Credential bundle | 5 MB of sign-in files; larger is not kept |
Accounts
| Session | 7 days, then sign in again |
| Waitlist form | 5 sign-ups per minute from one address |
Model providers have limits of their own - requests per minute, tokens per day,
and for subscriptions a usage allowance per few hours or per week. An agent that
hits one shows the provider's message in its terminal, and usually stops with
status error. See the FAQ.