The Open Anonymity Project – frontier AI models with privacy

The Open Anonymity Project launched a service that routes queries to open and closed frontier AI models through unlinkable inference, so model providers receive prompts without knowing the user's identity, according to the project's site. The service runs a public verification code in a Trusted Execution Environment (TEE) that users can run to confirm OA does not log queries, and adds a query rewriter implemented with a small language model in a GPU enclave, user-owned memory managed by memory agents in secure GPU enclaves, and an in-browser network proxy built with the Refraction Network. OA also offers Parallel and Council modes, prompt scrubbing, and personal memory agents, with code on GitHub and a blog post describing an Ethereum Foundation collaboration.

Access frontier AI models with your privacy in mind. Choose an option below to get started. The Open Anonymity Project zkAPI Usage Credits By continuing, you agree to our Terms https://chat.openanonymity.ai/terms and Privacy Policy https://chat.openanonymity.ai/privacy . Explore the code on GitHub https://github.com/OpenAnonymity , read our blog https://openanonymity.ai/blog/oa-ethereum-foundation-collab , or join our Discord community https://discord.gg/W5M6JMrfMz . Capabilities and privacy features - Unlinkable inference Unlinkable inference https://openanonymity.ai/blog/unlinkable-inference/ separates your queries from each other and from your identity. The Open Anonymity Project OA does not know which queries you send, and model providers receive them without knowing your identity. - Verifiable privacyUsing OA’s public verification code https://github.com/OpenAnonymity/oa-verifier , which runs in a Trusted Execution Environment TEE , you can confirm that OA does not log your queries. - Full-featured AI assistantOA provides access to open and closed AI models as well as unique features such as Parallel and Council modes https://github.com/karpathy/llm-council , prompt scrubbing, and personal memory agents. - Confidential query rewritingPrompts can reveal personally identifiable information or a distinctive writing style. To hide these details, OA comes with a query rewriter try tab-tab on your input implemented with a small LM running in a GPU enclave https://www.nvidia.com/en-us/data-center/solutions/confidential-computing/ . - User-owned memoryWith OA, your conversation history remains on your device. Memory agents https://openanonymity.ai/blog/nanomem/ running in secure GPU enclaves https://tinfoil.sh/ manage your memory and select relevant context to include in each prompt, without sharing your complete history. - In-browser network proxyNetwork-level metadata may also be revealing; for this reason, OA has partnered with the Refraction Network https://refraction.network/ to provide an in-browser network proxy.