Janus Subnet

Janus Subnet is a universal privacy pre-consensus layer designed for the Agent Economy. The core technology is based on the fusion of Confidential Computing (TEE) and decentralized, uncensored Large Language Models to create an "Asynchronous Digital Dark Room". Utilizing a dual-track heterogeneous architecture, we decouple Inner-layer Routing Nodes from Outer-layer Simulation Nodes , achieving physical-level data security where intent is "available but invisible". The project introduces a "Dynamic Poisson Honeypot" anti-cheat mechanism and Remote Attestation to ensure authentic compute contribution and intent consensus in a zero-trust environment.

Completed: Finalized the underlying architecture design and role definitions for the dual-track nodes (Track A/B). We have successfully executed the hardware-level Remote Attestation signature verification process on Netuid 359. Additionally, we completed VRAM overhead testing for the Inner-layer Matching Engine (14B Quantized) within the TEE environment and developed the initial intent desensitization logic.

Future Vision: We plan to conduct stress tests on the "Sword of Damocles" dynamic honeypot slashing logic on the testnet. Moving forward, we will fully release the APIs for the outer 72B simulation nodes and establish a USDC-based "Intent Gas" economic flywheel to attract top-tier global H100 compute, ultimately achieving a paradigm shift from "Asset Consensus" to "Probability and Intent Consensus".

The Janus Protocol has been in deep planning and research since last November, driven by a commitment to delivering a superior user experience that solves real-world frictions in the Agent Economy. In response to the rapid evolution of Confidential Computing and AI models, we proactively adjust our mechanism design to ensure a dynamic balance between security and utility. We firmly believe in the power of decentralized networks and warmly welcome all suggestions and feedback from community members and industry experts.