Umbra
Query your confidential documents securely. Upload sensitive files and ask questions inside a locked-down confidential workspace. Every interaction stays within a protected channel and runtime.
Confidential Chat with Cryptographic Guarantees
Umbra is a Confidential Chat that allows you to query documents with cryptographic guarantees. Your data is encrypted client-side in the browser, and the machine processing your queries is verified through cryptographic means. Umbra relies on Trusted Execution Environments (TEE) to ensure your sensitive documents remain private throughout the entire process.
End-to-End Protection
At each step of the process, the secure machine code and integrity are verified cryptographically. Your data never leaves the protected environment unencrypted.
Client Encryption
Data is encrypted in your browser before transmission
Secure Machine
Encrypted data reaches TEE with cryptographic verification
Decryption
Data is decrypted inside the secure TEE environment
AI Processing
Your documents are processed by AI within the secure environment
Encryption
Results are encrypted before leaving the TEE
Client
Encrypted response is sent back to your browser
Decryption
You decrypt and view the results locally
Security Assumptions: Umbra assumes the TEE hardware vendors (Intel, AMD, NVIDIA) are trusted and correctly implements the security guarantees. The cryptographic verification ensures that only verified code runs in the TEE, protecting against software-based attacks.
Building the Future of Confidential AI
We are a team of AI researchers, Security researchers, AI engineers, and Security engineers that seek to build the best solutions for confidentiality, privacy, and IP protection using state-of-the-art technology. Our team has deep expertise in TEE (Trusted Execution Environments), FHE (Fully Homomorphic Encryption), PPML (Privacy-Preserving Machine Learning), Side channels, and Hardware security.
Prof. Christina Garman
Expert in privacy, applied cryptography, and hardware. Co-Founder of Zcash, the first privacy-preserving cryptocurrency and large-scale deployment of zero-knowledge proofs.
Prof. Daniel Genkin
Expert in hardware security, trusted execution environments, and side-channel attacks. Co-discoverer of Spectre and Meltdown.
Dr. Jordan Frery
Leads the ML team at Zama, driving confidential-AI pipelines that keep data and models private using cutting-edge methods like homomorphic encryption. PhD in Machine Learning.
Launch a confidential AI program that scales with your compliance and trust requirements.
Partner with our security engineers to deploy in your preferred region, integrate with existing data controls, and evolve your policies alongside secure AI workloads.