hosted inference.
the provider runs it.
you trust its controls.
your request goes to their infrastructure.
you rely on the provider’s policies and assurances about how it is handled.
confidential AI.
verifiable by you.
we're building uk-hosted confidential inference,
because your data deserves more than a promise.
unlock the potential of your most sensitive data
vale lets you use AI with sensitive data on UK sovereign infrastructure. bring your own model, or use an open one. our drop-in SDK verifies where and how your request will be processed before your data is sent.
the vale SDK runs in your infrastructure. it checks CPU and GPU evidence, the approved software, and the server’s identity. if any check fails, your request stays with you.
your request travels over an encrypted connection that ends inside the verified environment. the network carrying it sees ciphertext, not your prompt.
the model uses your prompt in protected CPU and GPU memory. the host operating system does not hold the memory-encryption keys, so administrator access does not grant access to that protected memory.
the response returns to your application, then prompt and response payloads are cleared. only allowlisted operational metadata has a separate retention schedule.
the provider runs it.
you trust its controls.
your request goes to their infrastructure.
you rely on the provider’s policies and assurances about how it is handled.
you control it.
you're responsible for it.
keep the model on your own infrastructure.
the hardware, updates, and capacity are yours to manage.
we operate it.
your SDK verifies it.
the SDK checks the hardware, the software running on it, and the connection’s identity. your data doesn't leave your network unless every check passes.
processed in protected memory.
response returned. payload cleared.
interested? build with us
reach out and we'll discuss what your deployment needs.