The studio is open source and complete on its own. The paid editions exist for hardware and organizational controls — never to unlock a feature you already had.
Self-hosted
Freeforever
The complete studio, Apache 2.0. Everything below runs on your own hardware, with no account.
Hosted
$29per user / month
For when the local machine runs out. Same studio, GPU-backed runs, and a workspace your team shares.
GPU compute billed separately by the hour. Local runs stay free and unlimited.
Join the waitlistSelf-hosted or private cloud
Customannual
For regulated environments where data residency is the condition of the project existing at all.
| Capability | Open Source | Cloud | Enterprise |
|---|---|---|---|
| All vision, NLP, and LLM tasks | Included | Included | Included |
| Local training and inference | Included | Included | Included |
| Data stays on your machine | Included | Your choice | Included |
| macOS desktop app | Included | Included | Included |
| GPU-backed hosted runs | Not included | Included | Included |
| Shared team workspaces | Not included | Included | Included |
| SSO / SAML, RBAC | Not included | Not included | Included |
| Audit logs | Not included | Not included | Included |
| Air-gapped deployment | Not included | Not included | Included |
| Support | Community | Named contact + SLA |
Health, government, defense, and manufacturing QA teams don’t treat data residency as a preference. Orinth Enterprise is built for that constraint: air-gapped deployment, an internal model mirror, SSO and RBAC, and audit logs that record who ran what against which dataset version.
We work with a small number of design partners at a time so the deployment gets real attention. If your environment has constraints this page doesn’t cover, tell us what they are.
The boundary
Orinth produces model outputs for human review.
It does not diagnose, and it does not make autonomous or final decisions. In regulated settings, validation and regulatory obligations remain yours — the app is the instrument, not the authority.
No. The open-source studio is the whole product: every modality, every lifecycle stage, no feature flags and no seat limits. Cloud and Enterprise add infrastructure and organizational controls — hosted GPUs, shared workspaces, SSO, audit logs — that only mean something above a single machine. Nothing in the local workflow is gated.
The local app keeps working, because it never depended on us. Export your datasets and models from the hosted workspace and open them in the local studio — same formats, same app.
The core studio is Apache 2.0 and stays that way. If we ever ship a component under a different license, it will be a separate, clearly-labeled component — not a relicense of what you already have.
No. If you have a CUDA GPU or Apple silicon, Orinth detects it and uses it locally. Cloud is for teams whose training needs outgrow the hardware they own.
Yes, with one caveat: Hugging Face Hub imports and the chat web-search toggle need a network. Everything else — labeling, training, testing, inference, serving — runs with no connection. Enterprise deployments ship with an internal model mirror so Hub access isn't required either.
Orinth is a research and engineering instrument for measuring model behavior. It produces model outputs for human review and is not a diagnostic device — it does not diagnose, and it does not make autonomous or final decisions. Any use in a regulated setting is subject to your own validation and regulatory obligations.
Cloud and Enterprise pricing is indicative while both are in early access. Orinth is a research and engineering instrument for measuring model behavior. It produces model outputs for human review — it does not provide clinical diagnosis or autonomous decision-making.