Start a conversation.
Evaluating PRIOR and hit a wall? Have a research question? Just want to talk through the problem? You're welcome to reach out about any of it.
Stay informed
Benchmark results and product announcements. We send these rarely, and there's no filler.
Double opt-in: we email you a confirmation link first. Benchmarks and release notes only, with no sharing, and every email has one-click unsubscribe. See our Privacy Policy.
Try PRIOR
No need to ask for access. There's a free 30-day trial with the full capability set and no card required. If you self-host an open-weights model, you can start evaluating today.
Research inquiries, integration questions, or press. Reaching out directly is fine, and we usually reply within one business day.
Built for a specific kind of builder.
PRIOR is a control plane that runs inside the open-weights models you host yourself, steering or blocking unsafe or off-policy responses before they reach the user. To do that, it needs access to the model's inner workings as it runs (the internal activations that hosted APIs don't expose). So if you're routing everything through a closed API, PRIOR isn't your tool yet.
But if you're self-hosting Llama, Qwen, Gemma, Mistral, or a compatible architecture, and you're building something that has to behave reliably under adversarial conditions, pass a compliance audit, or hold a persona across a long conversation, we should talk.
The best first message is a simple one: here's what I'm building, here's where it's failing, and here's what I've already tried. We'll tell you honestly whether PRIOR can help.