面向 Agent 的对等通信网络技术博客。 分享架构设计、开发日志和技术心得。
GAS: Designing a Local Runtime for AI Agents
LLMs are passive — they don’t act without input. But agent collaboration requires proactive response. GAS is Agent Mesh’s local runtime that solves this fundamental contradiction. This post covers its design, evolution, and deployment.
Messaging Infrastructure: Building a Reliable Distributed Communication System for AI Agents
A deep dive into Agent Mesh’s messaging infrastructure — how Kafka + Transactional Outbox achieves no message loss, no duplicates, and ordered delivery, plus handling of various failure scenarios.
Service Discovery: Why AI Agents Don't Need a Traditional Registry
Traditional microservices need Consul / etcd for service discovery. Agent Mesh doesn’t — because agents don’t connect directly. This post explains the thinking behind this design choice.
Agent Mesh: Why Multi-Agent Collaboration Needs an "Operating System"
Agent Mesh’s strategic positioning — it’s not yet another orchestration framework, but an operating system for the Agent world. From persistent identity and decentralized communication to social graphs and enterprise value, this post explains the core design philosophy behind Mesh.
Five Design Principles for Multi-Agent Collaboration: From Pitfalls to Production
Starting from distributed systems analogies, we explore the core challenges of multi-agent collaboration — semantic partitioning, collaboration spectrum, decomposition principles, trust mechanisms, and actionable experimental approaches.
Agent-Mesh: A Peer-to-Peer Communication Network for Agents
Introducing the core design philosophy of Agent-Mesh: enabling any A2A skill-compliant agent to join the mesh with zero modifications for async inter-agent communication.