Hey Golang community,
You may have noticed that we’ve been posting a lot in this forum lately. Each of the libraries we’ve shared are components of our closed-source product, CONFSEC, and today, we’re excited to announce that we’re open-sourcing its framework as OpenPCC, the new standard for private AI inference.
What is OpenPCC?
OpenPCC is a Go-based framework for privacy-preserving AI inference engine. OpenPCC allows you to run open or custom LLMs without exposing prompts, outputs, or logs. It’s inspired by Apple’s PCC but fully open, auditable, and deployable on your own bare metal. The core idea is to layer privacy primitives between users and models: encrypted streaming, attested hardware, and unlinkable requests. No trust required; everything’s verifiable via transparency logs, and secured with TEEs, TPMs, blind signatures, and more.
It leverages several Go libraries we recently released:
- twoway: additive secret sharing & secure multiparty computation
- go-nvtrust: hardware attestation for NVIDIA H100 & Blackwell GPUs
- bhttp – Binary HTTP (RFC 9292) message encoding/decoding
- ohttp – request unlinkability to separate user identity from inference traffic
Why does this exist?
Current “private AI” solutions still require you to send sensitive inputs to model vendors. For anyone who cares about their privacy, that’s not acceptable. We built OpenPCC so anyone can deploy real private AI with cryptographic safeguards, backed by transparent Go code, to prevent any of their AI data from being stored, seen, or retained anywhere.
Key Capabilities
- Private LLM inference
- End to end encryption
- Confidential GPU verification with attestation
- Supports open models (Llama 3.1, Mistral, DeepSeek, etc.) and custom models
Get Started
Repository: https://github.com/openpcc/openpcc
White paper: https://raw.githubusercontent.com/openpcc/openpcc/main/whitepaper/openpcc.pdf
License: Apache-2.0
We’d love feedback, ideas, contributors, or security reviews from this community. If you’re building secure infrastructure or AI systems in Go, we hope this is useful to you.
Cheers,
The Confident Security Team