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

OpenTeam Details a Local Control Protocol for Browser AI Agents

AI summary

OpenTeam's author detailed a localhost control protocol that lets local agents such as Codex and Claude Code coordinate browser-based AI chats through openteamcli, a daemon, and a Chrome extension. The design uses bearer-token authentication, request IDs, timeouts, recoverable error codes, partial-result status, and exponential reconnection to handle the extension's intermittent lifecycle. The author also notes limitations including Chromium-only support, a single extension connection, local-only operation, reliance on DOM automation, and pending requests timing out rather than failing immediately after a disconnect.

In-depth insight

The protocol offers a concrete reference for securely exposing stateful browser AI sessions to local agents while giving failures enough structure for agents to recover.

上一篇写的是数据面:怎么把问题塞进网页输入框、怎么在没有 finish_reason 的世界里判断"回答完了"(把五个网页版 AI 拉进同一个群聊:不用 API key 的多 Agent 编排)。

这一篇写另一个方向,也是多模型群聊里更少人讨论的一半:控制面。

数据面解决"话怎么送进去、怎么取出来";控制面解决"谁有权让谁干活、失败了怎么自己爬起来"。当你希望调用方不是人、而是一个本机 agent(Codex、Claude Code 之类)的时候,控制面的设计决定了这件事能不能用。


一、先把拓扑摆出来

Agent(Codex / Claude Code)
   │  读 SKILL.md,执行 shell 命令
   ▼
openteamcli(进程,跑完就退出)
   │  HTTP POST /command  +  Authorization: Bearer <token>
   ▼
本地 daemon  ──  127.0.0.1:19305,只能本机访问
   │  WebSocket /ext
   ▼
Chrome 扩展 service worker
   │  runtime message
   ▼
各站点 iframe(ChatGPT / Claude / Gemini / DeepSeek / Grok)

#OpenTeam#openteamcli#Codex#Claude Code#ChatGPT#Claude#Gemini#DeepSeek#Grok#Chrome

Original · 11 min readRead the original
Group chat

No comments yet. Start the conversation.