Pick the one that fits how your organisation works
The first two decide where the system runs; the last two decide how you buy capacity. You can combine them: self-host one instance and still buy credit packs here for outside collaborators.
Deploy it once and get your own private dsh platform
The code is open source. Deploy it on your own servers or private cloud and you have a complete internal agent platform: accounts, quotas, the cloud workspace and the model gateway all sit inside your perimeter, and no data leaves your network.
- A single docker compose command brings up the whole stack
- Point the model upstream at your own — self-hosted, a dedicated gateway or a public cloud
- The gateway exposes an OpenAI-compatible API, so existing code connects by changing one base_url
- It can hook into your internal identity system, with usage accounted per department
Subscribe by seat, with credits and machine hours shared across the organisation
An admin buys seats and invites members. Everyone draws on one shared pool of credits and machine hours, and the console shows exactly who spent what. Nobody has to apply for their own model key or file their own expenses.
- Seats are billed monthly or yearly and can be added or removed at any time, from 3 seats up
- The shared pool stacks on top of personal balances, so joining a team never leaves a member with less
- You can cap an individual member so that one person cannot drain the whole team
- It fits handing out quota per person — teaching, research, outside collaborators: they get an account rather than a key, and it stops when the quota is gone
Buy credit packs and spend what you use
It works without a subscription. Credits are not wiped at the end of the month, which suits uneven usage or the occasional task. Every model is available — credits are the only gate, so there is no higher plan you have to buy before you can use a better model.
- A one-off purchase, and the balance stays valid long term
- Multipliers are benchmarked on Claude Sonnet at 1.00x, going as low as 0.044x
- Desktop Harness and Cloud Harness draw on the same credit balance
Pay monthly or yearly, with monthly credits and cloud machine hours included
The best value when your usage is steady. Credits are granted each month, and cloud workspace machine hours are granted separately — neither draws on the other, and machine hours count the time the workspace is running.
- Four plans — Free, Plus, Pro and Max — starting at 3 machine hours
- Yearly costs less than monthly, and you can move up or down a plan at any time
- Desktop Harness uses no machine hours, so heavy users can stay on the desktop and save them
Need a self-hosted deployment or team seats?
Tell us your scale and your use case and we will suggest a configuration.
Contact us