OpenPCC is an open-source framework designed for provably private AI inference, offering a transparent, auditable, and self-hosted alternative to proprietary privacy solutions. Inspired by Apple’s Private Cloud Compute, OpenPCC enables users to run open or custom AI models without exposing prompts, outputs, or logs. It enforces privacy through encrypted data streaming, hardware attestation, and unlinkable requests, ensuring that no single entity can trace or link user inputs to specific model executions. The framework is built to be community-driven and fully deployable on private infrastructure, making it ideal for organizations that require strong data privacy and compliance. OpenPCC aims to become a standard for AI data privacy, with a focus on transparency, verifiability, and user control. The full technical design is detailed in the OpenPCC Whitepaper, available at https://github.com/openpcc/openpcc/blob/main/whitepaper/openpcc.pdf. The OpenPCC Client repository provides a Go implementation that serves as the foundation for clients in other languages, including Python and JavaScript. It includes a set of in-memory services for testing and development purposes. Developers can use the client to make secure, privacy-preserving inference requests to OpenPCC-compliant servers. A basic example of how to use the client is shown in the cmd/test-client/main.go file. To connect to a production service, developers can configure the client with an API URL, API key, and identity verification policies. For instance, a request can be set up to target specific compute nodes by adding a X-Confsec-Node-Tags header, such as qwen3:1.7b, ensuring the model runs on a matching server. The client uses a custom HTTP request with a JSON body formatted for OpenAI-style API calls, such as generating a response to the prompt "why is the sky blue?". The request is then sent through the client’s RoundTrip method, which handles encryption, identity verification, and secure routing. For development, the project uses the Mage build tool. Running mage runMemServices starts all in-memory OpenPCC services locally, while mage runClient sends a test request through the system. Developers can also install Mage globally with go install github.com/magefile/mage@latest to streamline command execution. OpenPCC represents a significant step toward democratizing privacy-preserving AI, empowering users and organizations to leverage AI models without compromising sensitive data.