Self-host the workload.
Let us run the boring parts.
We're designing self-hosted OurAI with the customers who need it most. Conversations, inference, and your audit log stay inside your perimeter — never on our servers, never in our backups.
What you'll run. What we'll run.
Customer content lives only on your instance. The parts that don't touch content — policies, billing, updates — are a managed service so you don't have to run them.
Everything that touches customer content runs in your perimeter. Your hardware, your network, your operators.
- —Conversations — every prompt, every response
- —Inference — model calls on hardware you control
- —Audit log — full event stream, in your perimeter
- —Identity & access — connected to your SSO
The administrative surface — the parts that don't hold customer content, run by us so you don't have to.
- —Policy editor — author and version your guardrails
- —Subscription & licensing — billing, seats, entitlements
- —Usage analytics — aggregate counts, never content
- —Update channel — versioned releases, opt-in cadence
What self-hosted OurAI will feel like.
Self-hosting is only worth offering if it's done right. These are the principles we're building toward — the commitments that shape every design decision as we develop it.
Self-hosted and managed run the same codebase. We push new versions — you pull them when you're ready. No second-class tier, no forced updates, no perpetual feature gap.
All models are available — including US frontier. If you have the server capacity, configure a local open-weight model and your conversations never leave your infrastructure. Only aggregate metadata reaches us.
If you have a server ready and identity wired up, the path from installer to first prompt should fit in a day. We measure ourselves against this — every step that doesn't fit, we fix.
Help us design it.
Tell us about your environment and constraints — the customers who shape the product get the first builds.