Limits
Every plan offers Lite, Small, Medium, Large, and XL machines, from 256 MiB to 12 GiB RAM. Choose a size in the dashboard or send {"catalogId":"node","size":"medium"} to POST /containers. Omitting size selects Lite.
| Size | vCPU | RAM | Disk | Units/hour |
|---|---|---|---|---|
Lite · lite | 1/16 | 256 MiB | 2 GB | 1 |
Small · small | ½ | 4 GiB | 8 GB | 6 |
Medium · medium | 1 | 6 GiB | 12 GB | 10 |
Large · large | 2 | 8 GiB | 16 GB | 16 |
XL · xl | 4 | 12 GiB | 20 GB | 28 |
| Plan | Billing | Unit-hours/month | Concurrent units / containers | Starts/UTC month | Session / idle |
|---|---|---|---|---|---|
| Prepaid | $5 minimum top-up | $0.02/unit-hour | 128 / 100 | 10,000 | 24 hours / 30 minutes |
| Builder (legacy) | $5 | 250 | 28 / 5 | 1,000 | 1 hour / 10 minutes |
| Pro (legacy) | $180 | 9,000 | 128 / 100 | 10,000 | 24 hours / 30 minutes |
| Scale (legacy) | $999 | 50,000 | 640 / 500 | 100,000 | 72 hours / 60 minutes |
Usage and deadlines
One compute-unit hour runs one unit for one hour. Small uses 6 units/hour; XL uses 28. Running and provisioning time, including idle time, counts toward usage. Runtime is reserved before launch up to the session deadline; stopping releases unused reserved runtime after reconciliation. Requests rejected before reservation use no compute. Ambiguous or unreadable starts retain their reservation until reconciled.
Prepaid accounts purchase credit dollar for dollar with a $5 minimum top-up and a $5 default spending cap per UTC month. Compute costs $0.02 per compute-unit hour. Every start and renewal reserves funded runtime against the wallet and cap; only elapsed allocation, including startup and idle time, is consumed. There is no separate start fee. Purchased credit carries forward, and sessions can cross UTC month boundaries. New production services require funding and cap room for 24 hours of the full desired fleet. Existing legacy plans retain fixed allowances that reset each UTC month without rollover; their sessions also end at the UTC month boundary. Changing plans, canceling, or resubscribing does not reset existing usage. All sessions end when their allocated runtime expires.
Both the container ceiling and concurrent compute-unit ceiling apply, including pending starts. For example, Builder can run one XL or four Small machines; Pro can run four XL machines; Scale can run 22 XL machines. Each combination must also fit the available monthly runtime allowance.
Starts are shared across the account and reset at the start of each UTC month. Failed starts count. Stopping, changing plans, canceling, or resubscribing does not reset usage. A pending start counts toward concurrency; a request rejected at capacity does not reserve a start.
Sessions end at the hard deadline or the paid period recognized at creation, whichever comes first. Upgrades and renewals do not lengthen running sessions. Activity extends only the idle deadline; polling and SSH token issuance do not.
Downgrades and cancellation take effect at renewal. An effective downgrade can stop excess containers and shorten deadlines. Loss of paid access stops containers.
Operation limits
| Operation | Limit |
|---|---|
| HTTP execution | Four operations shared by foreground commands, managed jobs and files per container; 30-second default, 60-second maximum; 16 KiB command, 32 KiB request; 1 MiB combined output. |
| Managed execution | When advertised by capabilities: 15-minute maximum within the lease; 32 records retained for one hour; 1 MiB combined output; eight attached streams per container. |
| HTTP file transfer | 1 MiB per upload or download; 30 seconds; shares the command pool. |
| Interactive access | Four combined browser/SSH connections per container; ten live SSH tokens per account; tokens last at most 15 minutes. |
| Protected previews | When advertised by capabilities: ports 1024–65535; eight active grants and sixteen connections per container; TTL 60–3600 seconds, default 900, clipped to the lease; 15-second response-header timeout and 1 MiB per WebSocket frame. |
| Private Services (local prototype) | When networking.privateServices is true: 16 networks/account, 32 members/network, one network/generation; plain HTTP to NAME.internal, registered ports 1024–65535; 1 MiB request/response and 10-second timeout. Production support is not qualified. |
| API keys | Twenty per account; key names up to 80 characters. |
| Image builds | When enabled: default ten per UTC month, three saved images, 300-second deadline, 16 KiB Dockerfile, 512 KiB compressed context. Read returned limits. |
Current platform boundaries
Ordinary stop discards unsaved files; save a filesystem workspace before stopping. JavaScript and Python SDKs are published. Process/memory resume, arbitrary custom machine configurations, team seats, audit exports, advanced logs, and priority capacity are not currently available. Outbound internet access is enabled; customer-configurable egress restrictions and a guaranteed residency region are not offered.
Self-service plans have no uptime SLA or automatic service credits. Prepaid compute consumes purchased credit, subject to your monthly spending cap. There are no new monthly subscriptions, bulk bonuses, or purchase tiers. Existing legacy plans retain fixed monthly fees and compute allowances. See Terms and Status.