Herdr 开源 herdr-core:保活终端与跨机编排的编码智能体运行时
观察帖称 Herdr 开源 herdr-core 及基于它的桌面应用(Apache 2.0),强调终端与智能体会话在崩溃/重启后保活,并支持跨机器编排;同帖对比 Warp Oz 与开源 Campfire 等多代理并列工作区。
长跑编码智能体最怕笔记本休眠或终端崩溃后会话蒸发。观察者 stretchcloud 称 Herdr 开源了 herdr-core 及基于它的桌面应用(Apache 2.0):目标是在崩溃与重启后仍保持终端与智能体会话,并让智能体跨机器编排。作者将其视为近几个月该赛道的又一正式玩家,并点名 Warp 的 Oz(闭源、偏云,新增跨挽具记忆与自动编排,可并列跑 Claude Code、Codex 与 Warp 自有代理)。
同帖还介绍开源侧的 Campfire:在同一浏览器标签并列运行 Claude Code、Codex、Goose、Aider、OpenHands 等,各代理落在独立 git worktree;风险操作需多数同意、任一拒绝即阻断,并带 30 秒超时;支持会话回放与「代理赛跑」对比结果。观点是:不必在「纯终端 vs GUI」二选一,Campfire 保留 PTY 与 Monaco 编辑器,只是不再手动照看五六个终端标签。
对软件研发智能化,这反映「多代理运行时 / 会话持久化」正在产品化。原帖为观察评论,未给出 Herdr 的基准数据、安全模型细节或生产采用名单;Campfire 功能描述亦来自该帖归纳。选型前应核对各自仓库许可证、默认权限模型与遥测开关。细节以 Herdr / Campfire 上游仓库为准。
对读者而言,可把本稿当作窗内一手动态的结构化摘要:关键数字、产品名与限定语均来自原帖或报道,未另行推断未披露的定价、配额或合规结论;若需上线决策,请回到厂商文档与控制台核对当前区域可用性。
Herdr open-sources herdr-core agent runtime
Another coding agent runtime landed open source on my timeline today. Herdr shipped herdr-core plus a desktop app built on it, Apache 2.0, built to keep terminals and agent sessions alive through crashes and restarts while letting agents orchestrate across machines.
That's the third serious entrant into this lane in the last few months, next to Warp's Oz, which just added cross harness memory and automatic orchestration for running Claude Code, Codex and Warp's own agent side by side, closed and cloud first. The pattern across all of them is the same bet: nobody wants to babysit one terminal tab per agent anymore, and nobody wants to lose a long running session because their laptop slept.
Campfire takes the open source side of that bet further. Run Claude Code, Codex, Goose, Aider and OpenHands next to each other in one browser tab, each one isolated in its own git worktree so they can't step on each other's changes. Permission voting means a risky action needs majority approval, or any single deny blocks it, with a 30 second timer so nothing stalls waiting on a human who stepped away. Session replay and agent races let you run the same task across backends and actually compare what came back, instead of trusting whichever agent you happened to reach for that day.
The terminal versus GUI argument around coding agents misses the point. You don't have to choose. Campfire keeps the PTY terminal and a Monaco editor right there, it just stops making you manage five terminal tabs and a mental model of which agent is doing what.