Open Source Implementation of Apple's Private Compute Cloud (github.com)

OpenPCC is an open-source reimplementation of Apple’s Private Cloud Compute designed to enable provably private AI inference you can run on your own infrastructure. It enforces privacy guarantees — encrypted streaming, hardware attestation, unlinkable requests and transparency logs/verifiers — so prompts, outputs and logs aren’t exposed to untrusted operators. The project aims to be a transparent, community‑governed standard for AI data privacy and ships a Go client, a C core library (for Python/JS bindings) and in‑memory services for local testing; a full whitepaper and examples live in the repo. Technically, OpenPCC integrates attestation and a transparency verifier/identity policy (e.g., OIDC subject/issuer regexes) to ensure compute nodes run expected code and identities match policy. Clients speak an OpenAI-style generate API and can tag requests (X-Confsec-Node-Tags) to route to nodes running specific models. The repo includes a runnable dev flow (mage targets to run mem services and a test client) and code showing how to create a configured client, post inference requests, and verify transparency. For the AI/ML community this lowers barriers to reproducible, auditable private inference, reduces vendor lock‑in, and provides a practical reference for privacy‑preserving deployments in regulated or sensitive-data contexts.