Small
A quiet desk for one long-running agent.
- Persistent workspace
- uv / Pixi installs kept
- MCP connector
- Fixed rate, no burst multiplier
Three sizes. One clock. Illustrative launch pricing — labelled as such until we have a rate card we will honour.
Stateful Python execution infrastructure for AI applications.
A quiet desk for one long-running agent.
The default. Packages stay put. Agents spawn children.
Heavier jobs that still need a known monthly number.
| Vendor | Shape of the bill | Public unit (Sep 2026) |
|---|---|---|
| Modal Sandboxes | Per-second CPU + memory | US$0.00003942 / core-sec, US$0.00000667 / GiB-sec (source) |
| E2B | Plan + usage; Pro 24h cap | US$0.000014 / vCPU-sec, US$0.0000045 / GiB-sec (source) |
| Northflank | Allocated vCPU + memory + disk | US$0.01667 / vCPU-hr, US$0.00833 / GB-hr (source) |
| PySmith Medium | Fixed $/hr while up | US$0.08 / hr, illustrative |
Benchmarks, not a promise we are cheaper. At low duty cycle, per-second wins. At continuous runtime, allocated and fixed SKUs are the comparable set.
No. These are illustrative launch SKUs so you can budget the shape. We will not invent a discount, a free-tier minute count, or a ‘typical customer saves X%’ banner.
While the instance exists. Busy or idle. If your agent is mostly off, a per-second sandbox is cheaper. See the desk post for a labelled comparison.
Workspace disk is included at SKU quota in preview thinking; extras will be itemised when they exist. We are not printing fake GB prices.
Not in v1. Pause is the competing product. Our wedge is the process that should not pause.
Email hello@pysmith.com with the SKU you would actually run. We would rather hear duty cycle than a wish-list.