Vagus — A Vagal-Nerve Layer for Safer Agents

Vagus 在 LLM/Planner 与执行器之间引入链上"迷走神经层",用传入证据(Afferent)、张力(VTI)与三态(ANS)、制动(Vagal Brake)、反射弧(Reflex)、短时效能力令牌(Capability Token) 来防止不对齐与危险动作。

Core Features

Three fundamental pillars that make Vagus a comprehensive safety layer for autonomous agents

🛡️

Safety Layer

Afferent evidence collection, tension assessment, and tri-state management (SAFE/DANGER/SHUTDOWN) to build neural-system-like safety mechanisms ensuring stable operation of agent systems.

Agent Protection

Vagal Brake, Reflex Arc, and Capability Token authorization work together to prevent dangerous actions and provide instant responsive protection.

🔧

Multi-Language Stack

Seamless integration of Rust WASM smart contracts, Rust device gateway, and Python planner, supporting complete security lifecycle management from blockchain to edge devices.

Project Components

Modular architecture designed for scalability and interoperability across the entire safety stack

📄

contracts/

Rust WASM smart contract core, including key components such as ans_state_manager, capability_issuer, vagal_brake, reflex_arc.

View Details
🌐

gateway/

Rust device-side gateway handling event monitoring, AEP reporting, VTI calculation to ensure secure communication between executors and blockchain.

View Details
🔮

oracle/

Rust Tone Oracle service calculating VTI and updating ANS state, providing decentralized tension assessment mechanism.

View Details
🤖

planner/

Python tools for intent generation, Schema parsing, EIP-712 signing, supporting secure execution of complex agent tasks.

View Details
📋

schemas/

YAML/JSON schemas & policies defining safety rules and action specifications for different executors (such as robotic arms).

View Details
📚

docs/

Complete documentation including VagusSpec.md, Architecture.md, ThreatModel.md, providing comprehensive technical reference.

View Documentation