Encrypted Vault Overview
An Encrypted Vault adds end-to-end encryption and confidential computing to standard vaults, so your prompts, responses, and everything in the middle are never exposed in plaintext, not even to Cohere or the cloud operator. Everything stays encrypted end to end inside hardware-backed trusted execution environments (TEEs), and you can cryptographically verify exactly what’s running before you send a single byte of data.
It gives you private AI without self-hosting: a one-click, fully managed SaaS experience with the privacy posture of an on-prem deployment, and privacy you can verify rather than take on trust.
Encrypted vaults deploy through the same shared vault flow as any other vault, and you call them with the same API. The one addition is the Cohere OHTTP proxy, a confidential-computing component that runs on the client side and establishes an attested, end-to-end-encrypted tunnel to the vault. It comes as a standalone proxy running in your environment or as a Python package that wraps your client’s transport layer. Either form verifies the deployment’s attestation and encrypts every request before any data leaves your machine, so your application code stays unchanged. See Using the API with an Encrypted Vault.
This section covers only what’s specific to encrypted vaults: the security model, verification, key management, compliance, and the models and pricing that apply.
What Encrypted Vaults add
- Data is protected in transit, at rest, and in use, so prompts and responses are never exposed in plaintext, not even to Cohere or the cloud operator.
- Inference runs inside hardware-backed TEEs spanning the CPU (a confidential VM such as Intel TDX or AMD SEV-SNP) and the GPU (an NVIDIA GPU in confidential-computing mode), and your requests stay encrypted from your client all the way into the TEE, so the load balancer and network in between never see plaintext.
- Verifiable remote attestation lets you confirm you are talking to a genuine TEE running the exact code you expect. (This is what the vault’s verification badge in the Model Vault app reflects.)
- Every inference response includes an attestation certificate you can check, so you can verify the TEE and its code on each request.
- Designed to help regulated organizations meet compliance obligations (for example, GDPR, HIPAA, SOC 2).
Don’t need confidential-computing protections? A Standard Vault offers the same managed, single-tenant deployment without the encryption-in-use and attestation features.
Deploy an Encrypted Vault
Follow the shared flow, choosing the Encrypted option when you create the vault:
Understand what Encrypted Vaults protect
Review the Security Model to understand the trust boundary and threat model, specifically who is (and is not) able to access your data.
Learn the foundations (optional)
New to confidential computing? Read the Confidential Computing Primer for a quick grounding in TEEs, secure enclaves, and GPU confidential computing.
Understand how verification works
You do not have to verify manually before each request: the Cohere OHTTP proxy checks the deployment’s attestation automatically and refuses the connection if it does not pass. To understand what is checked, and how you can inspect the evidence yourself, see Remote Attestation and Verifying Your Deployment.
Configure encryption and keys
Review Encryption & Key Management to understand how data is protected in transit, at rest, and in use, and how keys (and ZDR) are handled.
Call your vault
Make inference requests using your vault endpoint. See Using the API with an Encrypted Vault, where the Cohere OHTTP proxy verifies the deployment’s attestation and encrypts every request before any data leaves your machine.
Confirm compliance
Check Compliance for how Encrypted Vaults map to GDPR, HIPAA, SOC 2, and audit-logging requirements.