Run all of it on your own metal
Uncensored companions, voice and generation inside your own infrastructure — no provider policy above yours, and no user data crossing a boundary you do not control.
Capacity, control and a direct line
The stack inside your perimeter
Every model already runs in a container. Enterprise means running those containers on your hardware, in your network, under your change control.
Self-hosted deploymentOrganisations and roles
Group API keys under an organisation with roles and a shared allowance, so a team is not passing one key around with no attribution.
Shared quotaAttribution per key
Usage is recorded per key and per endpoint, so you can charge a cost centre or a customer rather than guessing at the split.
Per-key analyticsEmbedded sessions
Publish a companion into a LiveKit room from your own backend and hand the browser a short-lived token. Your secret key never leaves your server.
WebRTC · session tokensSigned webhook delivery
Seven production events, each signed with a timestamped HMAC over the body, with a replay window, constant-time comparison and retries on failure.
7 events · HMAC-signedDirect engineering contact
You talk to the people who wrote it. At this stage that is a genuine advantage over a ticket queue, and it is what we can actually offer.
No support tier theatre
- Authenticatedscoped key · sk_live_
- Meteredquota enforced per key
- Inferredcontainers you operate
- Attributedper key · per endpoint
Every call answers to a key
Keys carry scopes and quotas, quotas are enforced on every API route rather than only at the edge, and usage lands in the analytics attributed per key and per endpoint.
- Scoped, quota-limited keys. A key does what it was issued to do, and no more than its allowance.
- Organisations and roles. Team access under a shared allowance instead of one credential passed around.
- Attribution you can bill against. Per key, per endpoint — a cost centre or a customer, not a guess.
- Live probes across the stack. Thirteen services, health-checked continuously.
The probes publish to the status page — the same view we watch, not a marketing uptime badge.
The answers you are about to ask for
Including the ones that are not yet a yes. You would find them in a security review regardless; better that you find them here.
| Question | Status | Answer |
|---|---|---|
| Where does inference run? | Confirmed | In containers you operate. No third-party AI provider receives prompts, audio or images. |
| Authentication | Confirmed | Sessions plus scoped API keys. Every mounted route is auth-gated, with automated tests asserting it. |
| Webhook authenticity | Confirmed | Timestamped HMAC bound to the body, a replay window, and constant-time comparison. |
| Rate limiting and quotas | Confirmed | Enforced per key on every API route, not only at the edge. |
| Production webhook events | Confirmed | All seven declared events emit from production paths, with a test asserting no declared event lacks a producer. |
| Age verification and moderation | Not built | Not shipped. You provide these in your application layer — see the compliance split. |
| GDPR erasure | Confirmed | Deleting an account unlinks the user’s stored files and removes their records. Every path is containment-checked before deletion, and a dry run is available. |
| SOC 2 / ISO 27001 | Not held | No certification is held. We will not claim one to win a procurement round. |
| Uptime SLA | Negotiable | No standard SLA is published. For a self-hosted deployment the availability is largely yours; terms can be agreed in a contract. |
| Payment and invoicing | Manual | No processor is connected. Enterprise agreements are arranged and invoiced directly. |
Before the first call
What does self-hosted actually mean here?
The containers that run every model — speech, language, synthesis, image and video — deployed on your hardware, in your network, under your change control. Inference never crosses a boundary you do not operate, because there is no hosted inference tier behind it.
How is Enterprise priced?
The listed plan is $299 a month with unlimited sessions, duration and API calls. No payment processor is connected, so agreements are arranged and invoiced directly — which also means terms can be shaped to the deployment rather than a checkout page.
How does a team get access?
Through an organisation: members carry roles, keys are grouped under a shared allowance, and usage is recorded per key and per endpoint. Nobody passes one credential around with no attribution.
How do we monitor a deployment?
The platform probes its thirteen services live, and the status page shows that view — the same one we look at. In a self-hosted deployment those probes run inside your perimeter with everything else.
Can we evaluate before signing anything?
Yes. The free tier needs no card and reaches every capability, the status page is public, and the diligence table above states the gaps as plainly as the strengths. Bring the answers it does not cover to a call.
Start the conversation
Tell us the volume and the constraints. We will tell you honestly whether we can meet them today.