oh-my-opencode-slim plugin
An OpenCode plugin: an orchestrator hands work to six specialists, each on its own model, and merges their results into one answer
- 9,844 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Agents read and edit project files and run commands on behalf of OpenCode, with background specialists working in parallel
- Code and questions go to several model providers at once when the council is used
- Built-in MCP servers call external documentation and GitHub code search services
Install
- Works in
Manual install
bunx oh-my-opencode-slim@latest installIf Bun is not installed, npx oh-my-opencode-slim@latest install works too. To make the OpenCode Go preset active, add the --preset=opencode-go flag.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents read and edit project files and run commands on behalf of OpenCode, with background specialists working in parallel
- Code and questions go to several model providers at once when the council is used
- Built-in MCP servers call external documentation and GitHub code search services
This is third-party code. Review the repository files before installing.
What it does
oh-my-opencode-slim adds a team of agents to OpenCode under an orchestrator. The orchestrator plans the work, sends specialists off to run in the background and merges their results before it continues. There are seven agents: an orchestrator, a codebase scout, an adviser for architecture and hard debugging, a council of several models, an external-documentation lookup, a UI implementer and a fast fixer. The config sets a separate model for each, from any provider, so an expensive model plans while a cheap one scouts. Install with `bunx oh-my-opencode-slim@latest install` or through `npx`. The installer adds the plugin to the `plugin` array in `~/.config/opencode/opencode.json` and creates `~/.config/opencode/oh-my-opencode-slim.json` with two model sets: OpenAI and OpenCode Go. The `/preset` command swaps the team's models on the fly. Permissions are granted by lists: each agent can be given its own skills and MCP servers. Eight skills are built in, among them `codemap` for a repository map, `deepwork` for a step-by-step mode on large tasks, `verification-planning`, `simplify`, `worktrees`, `clonedeps` and `reflect`. There are LSP tools, syntax-tree search for 25 languages, built-in MCP servers for documentation and GitHub code search, live views of agent work in Tmux, Zellij, Herdr, cmux, kitty or WezTerm panes, and an optional floating Companion window.
Who it is for. For OpenCode users who want to split tasks across several models and keep costs under control.
Good fit when
- One model handles the plan, the code search and the edits, and you do not want to pay for a strong model on every step
- A task splits into independent parts that specialists can do in parallel in the background
- You want advice from several models at once on a disputed architecture question
- You need different models for different roles: a strong one for planning, a fast one for scouting, a multimodal one for UI
- You want to watch the agents work in separate terminal panes
Not a fit when
- You do not work in OpenCode: the plugin is written for it alone
- You want a single model with no delegation: background agents multiply the number of requests and the cost
- You are looking for an orchestrator plugin for OpenCode and Codex CLI together: oh-my-openagent in the catalog is closer
- You want a council of models on every question: the author deliberately does not call it automatically, since it queries several models at once
Example request
Have the scout find every place where we read settings from the environment, then have the fixer move them into one module and run the testsLimitations
The license is MIT. The plugin works only inside OpenCode and calls itself a slimmed-down fork of oh-my-opencode.
- You need provider logins through `opencode auth login`, and the models in the config have to be matched to your own subscriptions: the author warns the ready-made sets are often out of date.
- The council agent queries several models in parallel, and background specialists run by default, so token use grows.
- Terminal panes need Tmux, Zellij, Herdr, cmux, kitty or WezTerm running; on OpenCode v2 hosts in `--standalone` mode they do not open.
- The Companion window is written in Rust and optional: the installer answers "no" by default.
- Changes to the model set apply to new child tasks, while edits to prompts, tools and permissions need a full OpenCode reload.
How to disable. Remove the plugin from the `plugin` array in `~/.config/opencode/opencode.json` and restart OpenCode. The `~/.config/opencode/oh-my-opencode-slim.json` settings file can be deleted separately. To turn off only part of it, use the config switches for tools, MCP servers, agents, skills, hooks and slash commands.
FAQ
Why seven agents instead of one?
Each gets its own model and its own set of permissions. The orchestrator needs a strong model for planning, the scout and the fixer do well on fast, cheap ones, and the council and the architecture adviser need the deepest reasoning.
How do I change the models for the whole team at once?
With `/preset`. Presets can inherit from a base one with `extends`: a `design` preset can take `base` and override only the UI implementer's model.
Why does the orchestrator not call the council by itself?
It is the most expensive path in the system, so the trigger is deliberately strict. Mention `council` or `@council` in your request and the procedure is injected once per session.
What does the installer do to my files?
It adds the plugin to `~/.config/opencode/opencode.json` and creates the plugin's settings file. Bundled skills are registered at load and not copied to disk; a same-named folder in `~/.config/opencode/skills/` overrides the bundled version.
Is there an extra Observer agent?
Yes, an optional one: the OpenCode Go preset enables it for image analysis because the orchestrator model in that preset is not multimodal.
Can I install third-party agents?
Yes, through marketplace packages. `oh-my-opencode-slim marketplace install author/package` installs a package and `enable` turns it on in the active preset. The new agent appears only after OpenCode reloads.
Discussion
No questions or reviews yet.
Related
PluginsOrchestrator plugin for OpenCode and Codex CLI: multiple agents, hooks, built-in MCPs, ultrawork mode and hash-anchored edits
An open source coding agent for the terminal and desktop with build and plan modes
Multi-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
An orchestration plugin for Claude Code, Codex, Gemini CLI and Qwen Code: 39 specialists, parallel subagents, persistent sessions and standalone audits
Memory for coding agents in plain Markdown: the agent keeps specs, decisions and lessons itself, and each new session starts from them. No vector database
A Claude Code plugin that runs one task across several external models and surfaces disagreements before you ship
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| oh-my-opencode-slim (this entry) | no | 9,844 | |
| oh-my-openagent | no | 70,440 | an orchestrator for OpenCode and Codex CLI with hooks and line-hash edits |
| OpenCode Swarm | no | 514 | OpenCode plugin that turns a session into a team of agents led by an |
README in short
The description is in English, with translations into Chinese, Japanese, Korean and Persian. It first explains the idea: instead of one model for every part of the work, the plugin hands parts out by quality, speed and cost. Then come a feature list, user quotes, two install routes plus a run from master, a sample config with two presets and descriptions of the seven agents with recommended models. It closes with sections on skills and Companion and an index of the docs in `docs/`. The repository holds TypeScript source, a Rust part, tests, a security policy in `SECURITY.md` and an `.out-of-scope` folder with the author's decisions on what not to add.