oh-my-openagent

English | 简体中文 | 日本語 | Русский | 한국어 | Español | Français | Deutsch | Português (Brasil) | Italiano

[!NOTE] OmO 5.0: OmO ❤️ Pi Install it with curl -fsSL https://get.omo.dev/install.sh | bash. Memory system, CodeMode, Anthropic subscriptions, all covered. Just type "mass ulw" with your prompt - now you are the master of graph engineering. Multi-model ultracode, together with a better memory system. (The right panel is omo-herdr-dag.)

Sponsors These are our sponsors. They help keep OmO going as a personal side project.

	OpenGateway sponsors OmO. An OpenAI-compatible gateway that puts many model providers behind one API. Thank you for backing open source.

[!TIP] Be with us!

	Join our Discord community to connect with contributors and fellow OmO users.
	Updates for OmO used to be posted on my X account. Since it was mistakenly suspended, @justsisyphus now posts updates on my behalf.
	Follow @code-yeongyu on GitHub for more projects.

OmO is one omo command that turns your tokens into finished work: research across ten thousand sources, a deck people actually get, backends, frontends, code. It runs on senpi, our fork of pi, with everything below built in.

Install

curl -fsSL https://get.omo.dev/install.sh | bash
omo

On Windows, in PowerShell: irm https://get.omo.dev/install.ps1 | iex. The script installs the native omo binary for your OS and CPU into ~/.local/bin, checks it against the release checksums, and falls back to GitHub Releases when the mirror is unreachable. Add -s -- beta (or a version such as -s -- 5.1.1) after bash to pick another channel. Windows CMD, fixing your PATH, updating and uninstalling are covered at omo.dev/docs/install.

Prefer a package manager? bun add -g omo-ai (or npm i -g omo-ai) installs the same OmO. The package is omo-ai; the unrelated omo package on npm is someone else's.

Open your project, run omo, describe the job. That's the whole setup.

In a session, /thinking <level> sets how hard the model thinks (/thinking alone opens a picker), Shift+Tab cycles through the levels, and /efforts <level> sets the reasoning effort directly on models with effort grades. To pick up earlier work, /resume (or /sessions) opens your past sessions; from the shell, omo -c continues the last one and omo -r opens the picker. /help lists everything else.

Coming from the OpenCode edition or LazyCodex? Run omo setup once. It carries your provider keys, custom providers, MCP servers, skills and model picks across. omo doctor then shows which task categories your providers can run and what the old install left behind, and omo doctor --models shows which model each category and agent runs and in what order (keep one out everywhere with exclude_models). The full walkthrough is Migrating from OpenCode.

To update, run the install line again (a package-manager install updates with omo update). rm ~/.local/bin/omo removes the binary install; bun remove -g omo-ai (or npm uninstall -g omo-ai) removes the package install.

Why OmO

ultrawork. Skip the study. If a task feels hard, add ultrawork (or ulw) to your prompt. The agent reads the codebase before touching a line, plans, proves each step before moving on, checks the result on the real surface, and stops when it's done.

Hundreds of agents, one keyword. Type mass ulw and the whole job becomes a graph of agents: deep research over thousands of sources, a full machine-learning pipeline, a migration nobody wants to do. Each part runs on the model that fits it, in parallel, and gets checked before anything is called done.

Kibitzer. Your agent doesn't forget you, or your work. Memory lives in a git repository of markdown files. A small, cheap model sits beside the expensive main agent, reads what you've done before, and nudges the main agent before it repeats a mistake. You stop re-explaining yourself.

Mix models without thinking about it. Use the frontier models you love together with the cheap ones. Just sign in: /login covers Claude, ChatGPT, Kimi and GLM subscriptions. Every model ships with a system prompt tuned for it, so you never pick the wrong one.

High-quality skills, only the ones you need. Your model is smart, but it still doesn't know a lot about solving real-world problems. OmO ships the missing knowledge as skills, browser use included, and loads only what the task needs.

Everything else is crafted. Independent tool calls run together in one small program, so twenty reads are one round trip. Malformed tool arguments are repaired before they run. Waiting on a build or a deploy is a subscription, not polling. Long work keeps a record of what was checked, so a restarted session picks up where it stopped. Config, skills and prompts reload live.

Configuration

Opinionated defaults, adjustable if you insist. Your settings live in ~/.omo/omo.jsonc, and a project can add its own .omo/omo.jsonc; the closest file wins. Every key is in the configuration reference.

Docs

• omo.dev/docs
• Migrating from OpenCode
• Configuration reference
• Features
• Computer use (experimental)
• Ultrawork Manifesto

Reporting a bug

The fastest way to report a bug is /bug inside a session (available in the omo app). It lists what it includes and lets you see the exact report before you send it, then sends it to the OmO team: the recent engine log, your environment and config with secrets blanked, the session transcript, and identity lines - your device name and, when telemetry is on, an install id and a session hash, plus your account details (email, name, account, user and device ids, plan, device name) when you report from the desktop app, so we can reply. Everything but the identity lines goes through the upload redactor. GitHub issues stay open for anyone who cannot use the in-app report.

Reviews

"It made me cancel my Cursor subscription. Unbelievable things are happening in the open source community." - Arthur Guiot

"If Claude Code does in 7 days what a human does in 3 months, OmO does it in 1 hour."
- B, Quant Researcher

"You guys should pull this into core and recruit him. Seriously. It's really, really, really good."
- Henning Kilset

Author's Note

When I have to take the wheel back from an agent, that isn't teamwork. The system failed. OmO is built on that idea. You say what you want; the agent researches, plans, writes code you couldn't tell from a senior engineer's, verifies it, and keeps going until it's finished. It spends tokens when they buy real speed and routes the easy parts to cheap models.

I'm this project's most obsessive user. Which model has the sharpest logic? Who writes the best prose? What did everyone else ship this week? OmO is where those answers land. Have a better way? PRs welcome.

The goal is an agent you stop noticing. You flip the switch, the light turns on. The long version is the Ultrawork Manifesto.

No affiliation with any project or model mentioned. The Ultragoal and UltraQA ideas come from oh-my-codex, reimplemented from concept for OmO.

Loved by professionals at

Indent, makers of Spray (influencer marketing solution), vovushop (cross-border commerce platform) and vreview (AI commerce review marketing solution). Google. Microsoft. Vercel. ELESTYLE, makers of elepay (multi-mobile payment gateway) and OneQR (mobile application SaaS for cashless solutions). Deepgram.

English | 简体中文 | 日本語 | Русский | 한국어 | Español | Français | Deutsch | Português (Brasil) | Italiano

[!NOTE] OmO 测试版: OmO ❤️ Pi 使用 curl -fsSL https://get.omo.dev/install.sh | bash 安装。记忆系统、CodeMode、Anthropic 订阅,全部支持。 只需在提示词中输入 "mass ulw" - 你就是图工程的大师。多模型 ultracode,搭配更出色的记忆系统。(右侧面板是 omo-herdr-dag)

Sponsors 以下是我们的赞助商,支持着这个个人副业项目的持续开发。

	OpenGateway 赞助了 OmO。它是一个兼容 OpenAI 的网关,通过一个 API 即可使用多家模型提供商。感谢对开源的支持。

[!TIP] 加入我们！

	加入我们的 Discord 社区，与贡献者及其他 OmO 用户交流。
	关于 OmO 的更新过去发布在我的 X 账号上。 因为账号被意外停用，现在由 @justsisyphus 代为发布更新。
	在 GitHub 上关注 @code-yeongyu 获取更多项目信息。

OmO 就是一个 omo 命令，把你的 token 变成真正完成的工作：上万个来源的研究、一看就懂的演示、后端、前端、编程。它运行在 senpi 上，这是我们基于 pi 的分支，下面的一切都已内置。

安装

curl -fsSL https://get.omo.dev/install.sh | bash
omo

Windows 请在 PowerShell 中运行 irm https://get.omo.dev/install.ps1 | iex。脚本会为你的系统和 CPU 安装原生 omo 到 ~/.local/bin，并用发布的校验和验证。想用包管理器的话，bun add -g omo-ai(或 npm i -g omo-ai)安装的是同一个 OmO。包名是 omo-ai；npm 上那个无关的 omo 包属于别人。Windows CMD、PATH 修复、更新和卸载见 omo.dev/docs/install。

打开项目，运行 omo，说出要做的事。设置就这么多。

在会话里，/thinking <level> 设置模型思考的深度（只输入 /thinking 会打开选择器），Shift+Tab 依次切换级别，/efforts <level> 在有强度分级的模型上直接设置推理强度。想接着之前的工作，用 /resume（或 /sessions）打开过去的会话；在 shell 里，omo -c 继续上一个会话，omo -r 打开会话列表。其余命令都在 /help 里。

从 OpenCode 版或 LazyCodex 迁移过来？运行一次 omo setup。它会把你的提供商密钥、自定义提供商、MCP 服务器、技能和模型选择一并带过来。

之后运行 omo doctor，可以看到你的提供商能跑哪些任务类别，以及旧安装留下了什么。完整步骤见 从 OpenCode 迁移。

omo update 原地更新。bun remove -g omo-ai（或 npm uninstall -g omo-ai）即可卸载。

为什么选 OmO

ultrawork。 不必学习。觉得任务很难，就在提示词里加上 ultrawork（或 ulw）。

代理会先读代码库再动手，制定计划，每一步验证后再继续，在真实环境中检查结果，完成后停下。

让数百个代理为你工作。只需一个关键词。 输入 mass ulw，整件工作就变成一张代理图：横跨数千来源的深度研究、一整套机器学习流水线、没人想做的迁移。

每个部分由最合适的模型并行完成，说完成之前先经过检查。

Kibitzer。你的代理不会忘记你，也不会忘记你的工作。 记忆保存在一个由 markdown 文件组成的 git 仓库里。

一个小而便宜的模型坐在昂贵的主代理旁边，查看你做过的事，在主代理重蹈覆辙之前轻轻提醒。你不必反复解释自己。

不用纠结的多模型。 把你喜爱的前沿模型和便宜的模型混着用。登录即可：/login 支持 Claude、ChatGPT、Kimi 和 GLM 订阅。

每个模型都自带为它调好的系统提示词，你永远不会选错。

高质量技能，只给你需要的那一点。 你的模型很聪明，但解决现实问题时仍有很多不知道的事。

OmO 把缺失的知识做成技能，包括浏览器操作，只加载当前任务需要的部分。

其余的一切都经过精心打造。 彼此独立的工具调用放进一个小程序里一起执行，二十次读取只需一次往返。格式错误的工具参数在执行前被修好。

等待构建或部署是一种订阅，而不是轮询。长任务会记录检查过的内容，会话重启后从停下的地方继续。配置、技能、提示词实时重新加载。

配置

默认值自有主张，你坚持的话也可以调整。你的设置在 ~/.omo/omo.jsonc，项目也可以添加自己的 .omo/omo.jsonc；离得最近的文件优先。

所有配置项见 配置参考。

文档

• omo.dev/docs
• 从 OpenCode 迁移
• 配置参考
• 功能
• 电脑操作（实验性功能）
• Ultrawork 宣言

评价

"It made me cancel my Cursor subscription. Unbelievable things are happening in the open source community." - Arthur Guiot

"If Claude Code does in 7 days what a human does in 3 months, OmO does it in 1 hour."
- B, Quant Researcher

"You guys should pull this into core and recruit him. Seriously. It's really, really, really good."
- Henning Kilset

作者的话

如果我不得不从代理手里把方向盘拿回来，那不是团队协作，而是系统失败了。OmO 就建立在这个理念之上。

你说出想要什么；代理负责研究、规划，写出与资深工程师难以区分的代码，验证它，一直推进到完成。它在能换来真正速度时才花 token，把简单的部分交给便宜的模型。

我是这个项目最着迷的用户。哪个模型逻辑最敏锐？谁的文字写得最好？这周别人都发布了什么？这些答案最终都落在 OmO 里。有更好的办法？欢迎 PR。

目标是一个你不再察觉到的代理。按下开关，灯就亮了。完整版本见 Ultrawork 宣言。

与文中提到的任何项目或模型均无关联。Ultragoal 和 UltraQA 的思路来自 oh-my-codex，并按其概念为 OmO 重新实现。

这些公司的专业人士在用

Indent, makers of Spray (influencer marketing solution), vovushop (cross-border commerce platform) and vreview (AI commerce review marketing solution). Google. Microsoft. Vercel. ELESTYLE, makers of elepay (multi-mobile payment gateway) and OneQR (mobile application SaaS for cashless solutions). Deepgram.