Today, OpenGradient launched OpenGradient Chat, a generative AI assistant built on the opposite premise: the questions that matter most are exactly the ones you should be able to ask without attaching your name to them. Finally, an AI you can tell anything.
Privacy you can verify
Most assistants ask you to trust a privacy policy, a written promise to handle your data well. OpenGradient Chat replaces the promise with proof.
Privacy is enforced in the architecture, in three layers:
- Encrypted on your device. Messages are encrypted locally before they leave the browser, with keys that live only on your device.
- Routed through an Oblivious HTTP relay. The relay sees your IP but only ciphertext; the downstream gateway sees plaintext but never your IP. No single party can correlate identity with content.
- Processed inside a TEE-isolated gateway. Prompts are decrypted only inside a trusted execution environment with remote attestation. Memory is sealed, and the operator can’t read or log it.
Because the enclave is attested, you can verify these guarantees yourself rather than take OpenGradient’s word for them. OpenGradient never holds a message it could trace back to you. It is the same verifiable foundation as OpenGradient’s core network, now in an everyday assistant.
Every frontier model, in one app
Underneath the privacy, it is a full assistant. A single app reaches the frontier models (ChatGPT, Claude, Gemini, Grok, and ByteDance Seed, to begin with) and lets you switch mid-conversation or run two side by side. It also includes live web search, uncensored image generation, and file uploads for PDFs, documents, code, and images. Anyone who signs up receives 1,000 free credits.
What’s next
OpenGradient Chat starts with text and images, and the roadmap is broader. Dedicated image and video models are coming to the same workspace, with the same privacy guarantees, so the prompts you write to make a poster, a product mock, or a short clip stay just as unlinkable from your identity as the questions you ask today. New models will be added regularly, and everything lives in one app.
The foundation
OpenGradient is backed by a16z crypto, SV Angel, and the NVIDIA Inception Program, with angel participation from Illia Polosukhin, a co-creator of the Transformer architecture that today’s frontier models are built on.
“For two years the trade has been the same: you get a smarter assistant by giving up more of yourself. We think that is backwards,” said Matthew Wang, co-founder and CEO of OpenGradient. “The questions where AI is most useful are the ones people are most afraid to type. We built OpenGradient Chat so you never have to choose between getting help and keeping it to yourself.”
OpenGradient Chat is available now at https://chat.opengradient.ai/.
About OpenGradient
OpenGradient is the Network for Open Intelligence – a decentralized infrastructure network designed to host, execute, and verify AI models at scale. Rather than functioning as a standalone blockchain, OpenGradient operates as a specialized AI coprocessor. It enables other applications, blockchains, or agents to outsource computationally-heavy tasks to a dedicated network of specialized GPU and Trusted Execution Environment (TEE) nodes for computation. OpenGradient Chat is the consumer face of the same architecture, applying that trust-minimized approach to everyday AI use. The team includes engineers and researchers from Two Sigma, Palantir, Google, Coinbase, Meta, and Amazon.
Website: https://www.opengradient.ai/ | Chat: https://chat.opengradient.ai/ | Docs: docs.opengradient.ai
Media Contact
OpenGradient
[email protected]
https://www.opengradient.ai/
SOURCE OpenGradient
