看板
整个集群，一目了然每个会话都按需要你做什么来归类
会话会根据实时事实自动归位：智能体正在做什么、它的 Pull Request、CI 和审查。你只需打开需要你处理的，其余的不用管。
按事实归位
泳道由智能体、它的 PR、CI 和审查决定，而不是靠某个人记得去更新状态。
一个会话一张卡片
任务、智能体、分支和 Pull Request 集于一处。打开即可查看聊天、终端、变更文件和预览。
需要你的，排在最前
问题、失败的检查和修改请求都汇集到同一条泳道，不会有事情在你看不到的地方等着你。
orchestrator
待处理
4
优化首屏窗口边框的对齐
正在编辑文件3 分钟前
迭代中
2
确定预览窗口的最终标题栏尺寸
#322开启
8/44 通过46 分钟前
审查中
2
验证预览环境变量
#415审查中
部署检查运行中18 分钟前
可合并
1
在顶部导航中加入 Orchestrator Logo
#326已批准
合并 PR3 小时前
待处理
4
优化首屏窗口边框的对齐
landing/window-border-pass
正在编辑文件3 分钟前
移除过时的自动生成图标导入
cleanup/stale-icon-imports
正在删除文件14 分钟前
记录预览别名的归属
deploy/alias-ownership
正在撰写部署说明18 分钟前
确认下载按钮文案是否继续按平台区分
landing/platform-download-copy
正在修改文案1 小时前
迭代中
2
确定预览窗口的最终标题栏尺寸
landing/titlebar-metrics
#322开启
8/44 通过46 分钟前
接入首屏示意图的进度延迟
landing/progression-timing
#331开启
调整间隔抖动22 分钟前
审查中
2
验证预览环境变量
deploy/preview-env
#415审查中
部署检查运行中18 分钟前
在 hydration 前预加载 GitHub Star 数
landing/preload-stars
#324审查中
38/60 项测试3 条评论
检查已通过1 小时前
可合并
1
在顶部导航中加入 Orchestrator Logo
landing/topbar-ao-logo
#326已批准
暂无评论
合并 PR3 小时前
归档18
这些公司的工程师都在用
功能
编排智能体
描述想要的结果，编排智能体调度整个集群
每个项目都有一个主智能体。告诉它你想构建什么，它就会规划工作，为每个部分在独立的 worktree 中启动一个工作智能体，只在需要你时才回来找你。
反馈闭环
CI 失败，由对应的智能体修复审查评论会回到编写这段代码的会话
Orchestrator 会盯住它创建的每个 Pull Request。失败的检查、审查评论和合并冲突，都会连同修复所需的信息一起发给负责该分支的工作智能体，直到 PR 获得批准。
失败的检查
失败的检查、检查名称和日志链接，会直接发给负责该分支的智能体。
审查评论
来自人和审查智能体的评论，会按讨论逐条回到同一个会话。
智能体审查
让审查智能体在人工之前先读每个 PR，一键把发现的问题发回去。
智能体
使用你已经信任的智能体支持 38 款编程智能体，可在聊天中或其自带终端里使用
为每个任务选择智能体和模型。在结构化聊天中协作，或打开它的原生终端界面。无论哪种方式，任务、分支和反馈都会保持关联。
结构化聊天
计划、命令和文件编辑都在一条清晰易读的时间线上，需要审批的地方就地审批。
原生终端界面
在同一个 worktree 中打开智能体自带的 TUI，或让会话在聊天和终端之间切换。
可自行操控的浏览器
每个工作智能体都有自己的浏览器来预览应用，智能体还能亲自点击操作，检查自己的成果。
移动端
你走到哪，智能体集群就跟到哪用手机回复智能体，代码仍留在你的机器上
通过你的网络或 Tailscale 将移动应用与桌面端配对。智能体需要你时会收到通知，你可以在聊天中回复，或打开它的终端。
关键时刻才通知你
智能体提问、Pull Request 就绪或检查失败时，都会提醒你。
随时随地回复
无需打开笔记本，就能在聊天中回复、批准计划或开始新任务。
运行在你的桌面端
智能体和代码都留在你的机器上。手机通过 LAN 或 Tailscale 直接连接到它。
看看实际效果
价格
本地版免费开源。价格仅适用于云端。
Free
在你自己的机器上运行智能体。开源，无需账号。
$0/月
包含内容
- Claude Code、Codex、Cursor 等
- 编排智能体、工作智能体和 PR 审查
- 使用自有 API 密钥，不经中转
- 基于 Apache 2.0 开源
Starter
推荐适合个人开发者：合上笔记本后，智能体仍继续工作。
$20/月
同时运行 2 个8 个会话
包含内容
- 你睡觉时，智能体仍在工作
- 即开即用的虚拟机或 Coder 开发工作区
- 在云端处理 Pull Request、CI 修复和审查
- 同一网络内最多两台主机
- 1 个编排智能体，每周 20 小时会话时长
Pro
适合更大的智能体集群，分布在你的所有机器上。
$40/月
同时运行 10 个25 个会话
包含内容
- 包含 Starter 的全部功能
- 通过中继连接，主机数量不限
- 同时运行 5 个编排智能体
- 每周 80 小时会话时长，每 6 小时最多 24 小时
- 每月 1 次用量重置
Enterprise
适合需要规模、管控和支持的团队。
定制
同时运行 25+ 个100+ 个会话
包含内容
- 包含 Pro 的全部功能
- 10+ 个编排智能体，无会话时长上限
- 可使用自有 Coder 部署
- SSO 与团队管理
- 提供 SLA 保障的优先支持
真实场景
来自 Orchestrator 用户的真实反馈。
Our in house software factory, built on top of original @useOrchestrator has been delivering impressive results. Built on our code and architecture context, it helped us innovate 50x faster while increasing the quality bar substantially.
Buchi Reddy B@buchireddyOrchestrator.inc helped me in removing the friction of using multiple harnesses at the same time. I can just work in peace without ever worrying about running grok, omp, or opencode CLI separately. I get the best of all worlds!
Bhavit Sharmaexperimented with @useOrchestrator alongside my daily driver @orca_build this morning - Orchestrator - best for big ideas that need splitting into multiple PRs without having to babysit each one - Orca - best for tasks that need more oversight and frontend UI work going to keep testing this setup out but I like how Orchestrator eliminated the need for manually tracking individual tasks that stemmed from a big feature set across sessions It automatically did the orchestration work I had manually been doing in Orca. Well done devs
Noah@noahliskThere hasn’t been a day in the last two months when I opened another IDE or ran a coding agent in a terminal app. Orchestrator.inc really changes how you think about work. It’s a mindset shift you can’t go back from.
Pritom Mazumdar“Of every tool I tested, Orchestrator.inc is the one that pushes hardest past session management into autonomous PR handling. Agents fix CI failures, respond to review comments, and manage their own PR lifecycle without asking me to approve each edit.”
Augment Code9 Open-Source Agent Orchestrators for AI Coding (2026)Our in house software factory, built on top of original @useOrchestrator has been delivering impressive results. Built on our code and architecture context, it helped us innovate 50x faster while increasing the quality bar substantially.
Buchi Reddy B@buchireddyOrchestrator.inc helped me in removing the friction of using multiple harnesses at the same time. I can just work in peace without ever worrying about running grok, omp, or opencode CLI separately. I get the best of all worlds!
Bhavit Sharmaexperimented with @useOrchestrator alongside my daily driver @orca_build this morning - Orchestrator - best for big ideas that need splitting into multiple PRs without having to babysit each one - Orca - best for tasks that need more oversight and frontend UI work going to keep testing this setup out but I like how Orchestrator eliminated the need for manually tracking individual tasks that stemmed from a big feature set across sessions It automatically did the orchestration work I had manually been doing in Orca. Well done devs
Noah@noahliskThere hasn’t been a day in the last two months when I opened another IDE or ran a coding agent in a terminal app. Orchestrator.inc really changes how you think about work. It’s a mindset shift you can’t go back from.
Pritom Mazumdar“Of every tool I tested, Orchestrator.inc is the one that pushes hardest past session management into autonomous PR handling. Agents fix CI failures, respond to review comments, and manage their own PR lifecycle without asking me to approve each edit.”
Augment Code9 Open-Source Agent Orchestrators for AI Coding (2026)Our in house software factory, built on top of original @useOrchestrator has been delivering impressive results. Built on our code and architecture context, it helped us innovate 50x faster while increasing the quality bar substantially.
Buchi Reddy B@buchireddyOrchestrator.inc helped me in removing the friction of using multiple harnesses at the same time. I can just work in peace without ever worrying about running grok, omp, or opencode CLI separately. I get the best of all worlds!
Bhavit Sharmaexperimented with @useOrchestrator alongside my daily driver @orca_build this morning - Orchestrator - best for big ideas that need splitting into multiple PRs without having to babysit each one - Orca - best for tasks that need more oversight and frontend UI work going to keep testing this setup out but I like how Orchestrator eliminated the need for manually tracking individual tasks that stemmed from a big feature set across sessions It automatically did the orchestration work I had manually been doing in Orca. Well done devs
Noah@noahliskThere hasn’t been a day in the last two months when I opened another IDE or ran a coding agent in a terminal app. Orchestrator.inc really changes how you think about work. It’s a mindset shift you can’t go back from.
Pritom Mazumdar“Of every tool I tested, Orchestrator.inc is the one that pushes hardest past session management into autonomous PR handling. Agents fix CI failures, respond to review comments, and manage their own PR lifecycle without asking me to approve each edit.”
Augment Code9 Open-Source Agent Orchestrators for AI Coding (2026)Our in house software factory, built on top of original @useOrchestrator has been delivering impressive results. Built on our code and architecture context, it helped us innovate 50x faster while increasing the quality bar substantially.
Buchi Reddy B@buchireddyOrchestrator.inc helped me in removing the friction of using multiple harnesses at the same time. I can just work in peace without ever worrying about running grok, omp, or opencode CLI separately. I get the best of all worlds!
Bhavit Sharmaexperimented with @useOrchestrator alongside my daily driver @orca_build this morning - Orchestrator - best for big ideas that need splitting into multiple PRs without having to babysit each one - Orca - best for tasks that need more oversight and frontend UI work going to keep testing this setup out but I like how Orchestrator eliminated the need for manually tracking individual tasks that stemmed from a big feature set across sessions It automatically did the orchestration work I had manually been doing in Orca. Well done devs
Noah@noahliskThere hasn’t been a day in the last two months when I opened another IDE or ran a coding agent in a terminal app. Orchestrator.inc really changes how you think about work. It’s a mindset shift you can’t go back from.
Pritom Mazumdar“Of every tool I tested, Orchestrator.inc is the one that pushes hardest past session management into autonomous PR handling. Agents fix CI failures, respond to review comments, and manage their own PR lifecycle without asking me to approve each edit.”
Augment Code9 Open-Source Agent Orchestrators for AI Coding (2026)
Orchestrator.inc really changes the way you develop. The orchestrator and kanban have been a game changer. I’m no longer confused about what agent is doing what; scoping tasks and spawning them off has been a breeze.
Aditi Chauhanour dev team is using it @themoeapp - amazing tool
erik schützler@erikschutzlerToday I shipped 5 features for Renzaflow. While playing a game using @useOrchestrator Not "vibe coding." Planned, agentic development. Here's how it worked: I planned the work and defined the outcomes. Then I handed it off to Orchestrator.inc, my agent orchestrator. Orchestrator.inc broke the work into tasks, spawned workers, handled code review feedback, resolved merge conflicts, fixed CI failures, and carried each feature through to completion. I never touched a single PR. The shift isn't "let AI write your code." It's this: Don't babysit agents. Orchestrate them. You stay responsible for direction, priorities, and quality — the things that actually require judgment. Your agents handle execution, end to end — the things that don't. That's the unlock. Not faster typing. A different job entirely.
WHOMeee!@nikhil_achalewe used Orchestrator.inc (@useOrchestrator) to red team the learning loop, and (@TensorMux) + GLM-4.7-Flash for policy induction. the result: 27 → 8 human reviews after learning 29/29 tests passed E1–E8 passed 0 incorrect autonomous applications demo ↓ youtu.be/b12UIm7X8iM?is…
Miles@milesnotmoralezOrchestrator.inc really changes the way you develop. The orchestrator and kanban have been a game changer. I’m no longer confused about what agent is doing what; scoping tasks and spawning them off has been a breeze.
Aditi Chauhanour dev team is using it @themoeapp - amazing tool
erik schützler@erikschutzlerToday I shipped 5 features for Renzaflow. While playing a game using @useOrchestrator Not "vibe coding." Planned, agentic development. Here's how it worked: I planned the work and defined the outcomes. Then I handed it off to Orchestrator.inc, my agent orchestrator. Orchestrator.inc broke the work into tasks, spawned workers, handled code review feedback, resolved merge conflicts, fixed CI failures, and carried each feature through to completion. I never touched a single PR. The shift isn't "let AI write your code." It's this: Don't babysit agents. Orchestrate them. You stay responsible for direction, priorities, and quality — the things that actually require judgment. Your agents handle execution, end to end — the things that don't. That's the unlock. Not faster typing. A different job entirely.
WHOMeee!@nikhil_achalewe used Orchestrator.inc (@useOrchestrator) to red team the learning loop, and (@TensorMux) + GLM-4.7-Flash for policy induction. the result: 27 → 8 human reviews after learning 29/29 tests passed E1–E8 passed 0 incorrect autonomous applications demo ↓ youtu.be/b12UIm7X8iM?is…
Miles@milesnotmoralezOrchestrator.inc really changes the way you develop. The orchestrator and kanban have been a game changer. I’m no longer confused about what agent is doing what; scoping tasks and spawning them off has been a breeze.
Aditi Chauhanour dev team is using it @themoeapp - amazing tool
erik schützler@erikschutzlerToday I shipped 5 features for Renzaflow. While playing a game using @useOrchestrator Not "vibe coding." Planned, agentic development. Here's how it worked: I planned the work and defined the outcomes. Then I handed it off to Orchestrator.inc, my agent orchestrator. Orchestrator.inc broke the work into tasks, spawned workers, handled code review feedback, resolved merge conflicts, fixed CI failures, and carried each feature through to completion. I never touched a single PR. The shift isn't "let AI write your code." It's this: Don't babysit agents. Orchestrate them. You stay responsible for direction, priorities, and quality — the things that actually require judgment. Your agents handle execution, end to end — the things that don't. That's the unlock. Not faster typing. A different job entirely.
WHOMeee!@nikhil_achalewe used Orchestrator.inc (@useOrchestrator) to red team the learning loop, and (@TensorMux) + GLM-4.7-Flash for policy induction. the result: 27 → 8 human reviews after learning 29/29 tests passed E1–E8 passed 0 incorrect autonomous applications demo ↓ youtu.be/b12UIm7X8iM?is…
Miles@milesnotmoralezOrchestrator.inc really changes the way you develop. The orchestrator and kanban have been a game changer. I’m no longer confused about what agent is doing what; scoping tasks and spawning them off has been a breeze.
Aditi Chauhanour dev team is using it @themoeapp - amazing tool
erik schützler@erikschutzlerToday I shipped 5 features for Renzaflow. While playing a game using @useOrchestrator Not "vibe coding." Planned, agentic development. Here's how it worked: I planned the work and defined the outcomes. Then I handed it off to Orchestrator.inc, my agent orchestrator. Orchestrator.inc broke the work into tasks, spawned workers, handled code review feedback, resolved merge conflicts, fixed CI failures, and carried each feature through to completion. I never touched a single PR. The shift isn't "let AI write your code." It's this: Don't babysit agents. Orchestrate them. You stay responsible for direction, priorities, and quality — the things that actually require judgment. Your agents handle execution, end to end — the things that don't. That's the unlock. Not faster typing. A different job entirely.
WHOMeee!@nikhil_achalewe used Orchestrator.inc (@useOrchestrator) to red team the learning loop, and (@TensorMux) + GLM-4.7-Flash for policy induction. the result: 27 → 8 human reviews after learning 29/29 tests passed E1–E8 passed 0 incorrect autonomous applications demo ↓ youtu.be/b12UIm7X8iM?is…
Miles@milesnotmoralez
常见
问题
我已经在用 Cursor 这类 IDE，这适合我吗？
Orchestrator 的设计初衷就是配合你现有的工具。我们原生支持深度链接到 Cursor 等 IDE，你可以直接在 IDE 中打开工作区和文件。Orchestrator 位于各个工具之上：你喜欢用哪个智能体都行，工作流始终保持一致。
支持哪些 AI 编程智能体？
Orchestrator 支持 38 款编程智能体，包括 Claude Code、OpenAI Codex、Cursor、OpenCode、Aider、Goose、GitHub Copilot 和 Gemini CLI，并且可以按项目选择智能体。每个项目都可以单独指定自己的工作智能体、编排智能体和审查智能体。
并行智能体系统是如何运作的？
编排智能体负责规划工作并创建工作智能体，每个工作智能体都运行在独立的 Git worktree 中，因此可以同时在不同分支或功能上工作而互不冲突。Orchestrator 会盯住它们提交的每个 Pull Request：CI 失败、审查评论和合并冲突都会转回给负责该分支的智能体。内置的 AO 审查智能体可以审查 PR，并把修改意见发回给同一个工作智能体，而你只需在一块看板上掌握整个智能体集群。
Orchestrator.inc 和 Agent Orchestrator 是同一个产品吗？
是的。Orchestrator.inc 是 Agent Orchestrator（AO）的新名称。它仍是同一款开源应用和同一个 ao CLI，源代码位于 https://github.com/OrchestratorInc/agent-orchestrator。
Orchestrator 可以免费使用吗？
是的。本地应用基于 Apache-2.0 许可证免费开源，支持在任意受支持的运行框架上运行不限数量的本地智能体，并包含编排智能体、工作智能体和 PR 审查。智能体在你电脑上的 Git worktree 中工作，Orchestrator 不会代理你的 API 调用，在本地运行智能体也无需账号或信用卡。你的手机或其他电脑可以在同一网络下连接，也可以通过可选的托管 HTTPS 隧道从任何地方连接；Tailscale 无法在 iPhone 或 iPad 上使用，在 Android 上也尚未验证，目前还没有内置的设备中继。付费的 Cloud 方案每月 $20 起，额外提供云端项目以及在托管沙箱中运行的云端智能体。
智能体可以在云端运行吗？
可以。Orchestrator Cloud 现已可用：在桌面应用中登录 Orchestrator Cloud，选择“Clone to the cloud”来挑选一个 GitHub 仓库，即可在托管沙箱中运行工作会话和编排会话。在桌面端，你可以使用聊天、远程终端、文件检查器，以及该会话的 Pull Request 和审查。云端智能体包含在 Starter、Pro 和 Enterprise 方案中。本地项目无需 Cloud 也能照常使用。
可以使用我自己的 API 密钥吗？
可以。对于本地项目，Orchestrator 会直接使用你机器上已安装的编程智能体 CLI 及其现有登录状态，无论是 API 密钥还是服务商订阅。它不会代理 API 调用，因此用量由各服务商分别计费，成本完全由你掌控。
还有疑问？
阅读文档，查看详细的使用指南。