google / agentmesh
A Google-built P2P mesh network that gives AI agents verifiable identity, secure tool discovery, and A2A communication across any infrastructure.
星标趋势
AI 分析
项目摘要
Agent Mesh is a Go-based private P2P mesh network for AI agents from Google. It provides secure agent identity, service discovery, and tool/model sharing across distributed environments (laptops, containers, clusters, phones) with policy-based access control and OpenID Connect authentication.
为什么值得关注
It implements the A2A (Agent-to-Agent) protocol for agent interoperability with a strong security model — nothing is reachable by default, identity is tied to an identity provider, and access is governed by mesh policies evaluated on service names. It bridges MCP-based tool publishing with P2P networking, enabling agents to work across NAT boundaries.
优势
- Strong security-first architecture with identity-based access control and policy grants on service names rather than addresses
- Well-defined component separation: control plane (auth/credentials), router (relay/discovery), and node (local agent runtime)
- Active development with 20 releases in 6 months, CI, tests, and Apache 2.0 license
- Practical P2P mesh design with relay support for NAT traversal, enabling agents to communicate across diverse environments
局限性
- Relatively new project (created April 2026) with no Docker support and no examples yet
- Documentation cut off mid-sentence in README, suggesting incomplete documentation for some components
使用场景
- Building private agent-to-agent networks across teams, clusters, and cloud providers
- Enabling AI agents on edge devices or behind NAT to discover and use tools/services from other agents
- Creating secure multi-agent systems where tool access is governed by identity-based policies