AI UNDERDOGSDAILY PICK
AI UNDERDOGS
一个控制台，管十多个 AI
One cockpit. Dozens of AI agents
dotdevdotdev/agentwire-dev
同时开十几个 Claude Code 会话
不用再一个个切终端复制粘贴了
Juggling a dozen Claude Code sessions?
Stop switching panes and pasting prompts one by one
不是终端分屏那么简单
Not just tiled terminals
每个 Agent 跑在独立的 Git worktree 里
互相不撞车
有命令面板把所有操作收到键盘上
还有个调度器排队分配任务
这三层叠起来，才是真正的并行
Each agent runs in its own git worktree
— no collisions. A command palette keeps everything
on the keyboard. A scheduler queues tasks across
agents. Stack these three layers and parallelism actually
works
★ SIGNAL 1
每层都能单独用
Every layer works standalone
大多数工具做出来就是一整坨
AgentWire 不一样：worktree 隔离
命令面板、调度器
每块都能单独拿走用
这才是真有产品思维——不是堆功能
是拆模块
Most tools ship as a monolith. Here, worktree
isolation, the command palette, the scheduler — each
one works independently. That's product thinking: not feature-stacking
but module-building
★ SIGNAL 2
语音控制，他真做了
Voice control — he actually shipped it
看看最近的提交记录：holding notices behind proposals
anchoring at render
作者在解决语音输入和 Agent 输出之间的时序冲突
这不是录个 demo 就完事
是真在啃交互硬骨头
The latest commits tackle voice-UX timing — holding
notices behind proposals, anchoring at render. Not a
YouTube demo. The author is actually solving the
hard interaction gap between voice input and agent
output
他在 HN 上的标题就是一句话：「通过 tmux 和语音控制来编排多个 Claude Code 会话」
没有「革命性」「颠覆」这些词
就是一个 indie hacker 描述自己在做什么
His HN title says it plain: 'Orchestrating many
Claude Code sessions via tmux and voice-control.' No
buzzwords. Just one indie hacker describing what he's
building
AI UNDERDOGS
tmux + Claude Code，同时跑
tmux + Claude Code, all running in parallel
dotdevdotdev/agentwire-dev
关注 · 每天发现更多 AI 神作
github.com/dotdevdotdev/agentwire-dev