oh-my-opencode: What It Is and How to Install It

oh-my-opencode is a batteries-included plugin for the OpenCode coding agent that installs a pre-tuned team of 11 AI agents, 54+ lifecycle hooks, and built-in MCP servers with a single command. The project was renamed to oh-my-openagent in 2026, but the original npm package name still works, and the pitch is unchanged: it does for OpenCode what oh-my-zsh did for the zsh shell.

What Is oh-my-opencode?

oh-my-opencode is an open source agent harness created by YeonGyu Kim (GitHub handle code-yeongyu) in December 2025, now maintained under the Sisyphus Labs banner. The README makes the analogy explicit: if OpenCode is Debian or Arch, oh-my-opencode is Ubuntu. You install one plugin and get a fully configured multi-agent setup instead of assembling agents, hooks, and MCP servers by hand.

The project grew fast. It was created on December 3, 2025, and the GitHub repository sits at roughly 64,500 stars as of July 2026, with the latest release (v4.15.1) published on July 1, 2026. It now ships two editions: the full Ultimate Edition as an OpenCode plugin, and a Light Edition called LazyCodex that ports a subset of the hooks and tools to the OpenAI Codex CLI.

Quick Refresher: What OpenCode Is

OpenCode is an open source, MIT-licensed terminal coding agent built by Anomaly (the SST team) with more than 181,000 GitHub stars as of July 2026. It works with almost any LLM provider, runs as a terminal TUI with desktop and IDE companions, and is configured through opencode.json files. The official OpenCode docs cover the config system: a global file at ~/.config/opencode/opencode.json, per-project files, and .opencode/ directories for custom agents, commands, and plugins.

That flexibility is the point, and also the problem oh-my-opencode solves. Out of the box you choose your own models, wire up your own MCP servers, and write your own agents. It is closer to a kit than a finished product, in the same way Claude Code feels before you add plugins and skills.

How to Install oh-my-opencode

The install is one command, run with Bun: bunx oh-my-openagent install. It launches a terminal UI that registers the plugin in your opencode.json, walks through provider authentication (Anthropic, Gemini, GitHub Copilot, Z.ai, OpenCode Zen), and assigns models to each agent.

# Ultimate Edition (OpenCode plugin, requires Bun)
bunx oh-my-openagent install

# Light Edition for the OpenAI Codex CLI
npx lazycodex-ai install

# Both at once
bunx oh-my-openagent install --platform=both

Three practical notes from the README. First, the maintainer actually recommends pasting the installation guide URL into a coding agent and letting the agent run the setup, because humans tend to misconfigure it. Second, do not run bunx omo; the omo npm package is unrelated. Third, if something breaks, bunx oh-my-opencode doctor runs diagnostics, and uninstalling is just removing the entry from the plugin array in ~/.config/opencode/opencode.json.

What oh-my-opencode Actually Configures

oh-my-opencode installs three layers at once: an agent team, a tooling upgrade, and a set of enforcement hooks. The agent roster is the headline feature. Sisyphus is the main orchestrator, tuned for Claude Opus, Kimi K2.6, or GLM-5.1 and designed to keep working until a task is fully done. Hephaestus is an autonomous deep worker on GPT-5.5, Prometheus is an interview-style planner, Oracle handles architecture and debugging questions, Librarian searches docs and code, and Explore does fast codebase greps. Agents delegate by capability category (such as deep, quick, and visual-engineering) rather than by hardcoded model names, which the README calls the end of manual model juggling. If you want context on which models fit which tier, see our Claude model comparison.

The tooling layer includes three always-on MCP servers (Exa for web search, Context7 for library docs, and Grep.app for GitHub-wide code search), LSP integration for rename and go-to-definition, AST-Grep across 25 languages, and a tmux-based interactive terminal. The most interesting piece is Hashline, a hash-anchored edit tool that tags each line so stale edits get rejected; the README claims it raised one weak model's edit success rate from 6.7% to 68.3%, though that figure is self-reported by the project.

The enforcement layer is what makes it feel like a different product: a Todo Enforcer that pulls idle agents back to their task list, a Comment Checker that strips low-value AI-generated comments, a Ralph-style /ulw-loop that keeps iterating until everything passes, and an opt-in Team Mode that coordinates a lead agent with up to 8 parallel members. There is also a Claude Code compatibility layer, so existing Claude Code hooks, commands, skills, and MCP servers are claimed to work unchanged.

oh-my-opencode vs Plain OpenCode

Plain OpenCode gives you the harness; oh-my-opencode gives you a configured team. The table below summarizes what changes after the one-command install, based on the project README and the OpenCode configuration docs as of July 2026.

The tradeoff is control versus convention. If you want a minimal agent with exactly the tools you choose, plain OpenCode plus a few hand-picked MCP servers stays leaner and easier to debug. If you want a Claude Code-style full experience on subscription models from multiple vendors, the preconfigured route saves days of setup.

Why It Is Now Called oh-my-openagent

The repository was renamed from oh-my-opencode to oh-my-openagent during 2026 as the project expanded beyond OpenCode to multiple harnesses, including the Codex CLI. GitHub redirects the old URL automatically, and the npm package is dual-published under both names, so oh-my-opencode commands in older tutorials still resolve.

Part of the backstory is the January 2026 change where Anthropic began rejecting Claude Pro and Max OAuth tokens in third-party harnesses, which cut OpenCode users off from Claude subscription pricing. We covered the same enforcement wave against OpenClaw in this breakdown. The oh-my-opencode README claims the project itself triggered the block; whatever the cause, the practical result is that the project now leans on OpenAI, Kimi, and GLM subscriptions, and its Codex-focused Light Edition exists largely because of that shift. Early community discussion is preserved in the original Show HN thread from December 2025.

Limitations and Caveats

oh-my-opencode is powerful but opinionated, and it carries real caveats. Know these before installing:

• Not MIT licensed. It uses SUL-1.0, a fair-code style sustainable use license. Fine for internal use, but check the terms before building a commercial product on top of it.
• Telemetry is on by default. Anonymous PostHog telemetry (capped at one event per machine per day) can be disabled with "telemetry": false or OMO_DISABLE_POSTHOG=1.
• Bun is required for the Ultimate Edition, and the setup is complex enough that the README recommends letting an agent perform the install.
• Runtime-injected MCP servers do not appear in opencode mcp list, which can make debugging confusing.
• Fast-moving codebase. The project is mid-refactor toward a multi-harness architecture and had over 800 open issues as of July 2026. Expect breaking changes between versions.
• Token consumption. A harness that runs multiple agents in loops until completion will burn far more tokens than a single-agent session. Budget accordingly on metered plans.

Related Guides

• Claude Plugins: What They Are and How to Install Them
• Claude Code Templates: The Community Setup Toolkit
• Claude Code MCP: How to Add and Manage MCP Servers
• Best Claude Models in 2026 Compared