Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Lume A2A connects Lume apps to a distributed agent fleet

AI summary

In a first-person technical report, Lume A2A’s author describes a three-layer architecture comprising the Lume server, lume-core for language and peer semantics, and a React console that connects directly to running applications. The default fleet grew from the crm and scanner nodes to seven nodes, six console pages, and five external capabilities; five nodes expose complete A2A peers, while tsm-hub and cicdkit use signed Peer HTTP while their A2A endpoints remain unfinished. The report also says task.delegate sends execution work to Resolve-Lume’s sidecar and notes that A2A JSON-RPC endpoints lack authentication and require additional protections for public deployment.

In-depth insight

The report provides concrete implementation details for combining agent discovery, signed peer calls, a multi-application console, and delegated execution in one distributed collaboration architecture.

Lume A2A 实战:让 Lume 应用接入分布式 Agent 协作网络

先摆一组数字:起点是两个真实运行的后端节点(crm、scanner),后来默认 fleet 长成 7 个节点、6 个控制台页面、5 个对外能力。

关键在于:这套结构从一开始就不是为「那两个起步应用」量身定的,而是按同一套接入契约设计的——任何 lume 应用只要实现 Agent Card、/a2a 端点与 peer 语义,就能照这套三层结构并排接入;从 2 个涨到 7 个,加的只是目录里的条目,没有推倒重来。这里要避免一个误读:「接入」不等于零改造,上面三样是每个节点自己要补齐的,统一的只是接入方式。

图里最容易被看错的一点是:7 个节点并不等于 7 个同类端点。默认 Fleet 里 5 个是原生完整 A2A 对等端(Agent Card + POST /a2a),2 个(tsm-hub/cicdkit)经签名 Peer HTTP 接入、A2A 端点还在补齐;而 resolve-lume 虽是完整对等端,却同时是全 fleet 唯一的执行器。

这些数字背后是同一个问题——把真实运行的 lume 应用并排放在一起时,控制台该怎么组织?我花了不短的时间在这上面,最后收敛成一张三层结构:lume 作服务端、lume-core 提供语言与 lib/peer.lume、React 控制台直接对接应用。这篇文章按时间顺序记录我为什么这么切,以及后来它长成了什么样子。

#Lume A2A#Lume#lume-core#Resolve-Lume#crm#scanner#tsm-hub#cicdkit#lan#talentlens

Original · 18 min readRead the original
Group chat

No comments yet. Start the conversation.