alvinunreal/oh-my-opencode-slim

oh‑my‑opencode‑slim – an AI‑agent orchestration plugin for OpenCode

What it is – A plugin that turns OpenCode (a code‑centric LLM platform) into a tiny “team” of specialized AI agents. Instead of sending every request to a single model, the plugin’s Orchestrator breaks a task into subtasks and dispatches them to the most suitable agent – Explorer, Oracle, Librarian, Designer, Fixer, or Council – possibly mixing models from different providers. The agents run in the background, their results are reconciled, and the user sees a single coherent answer.

Why it matters – Large‑language‑model workflows often suffer from a “one‑size‑fits‑all” model that is either too slow, too cheap, or lacks the right expertise. By routing work to the right specialist, oh‑my‑opencode‑slim aims to improve:

• Quality – stronger reasoning models handle design/architecture, fast models handle scouting.
• Speed – parallel background agents let work proceed while the Orchestrator continues planning.
• Cost – cheap models are used for cheap tasks, expensive models only where needed.

---

Core concepts

Agent	Role	Typical model (default)
Orchestrator	Master planner, schedules and reconciles work	openai/gpt-6-sol (medium)
Explorer	Scans the codebase, builds a quick map	openai/gpt-6-luna
Oracle	Gives high‑level architectural advice, deep debugging	openai/gpt-6-astra (high)
Council	Runs several models in parallel and synthesises a consensus	Config‑driven – uses a strong synthesis model and a set of “councillor” models
Librarian	Retrieves external docs, web or GitHub knowledge	openai/gpt-6-luna
Designer	Produces UI/UX sketches or visual artefacts	openai/gpt-6-luna
Fixer	Performs focused implementation or bug‑fix tasks	openai/gpt-6-luna

Highlighted features (as described in the README)

• Background orchestration – agents run as independent tasks; the Orchestrator can wake‑up when results are ready.
• Prompt‑based skills – built‑in workflows such as deepwork, codemap, verification‑planning, reflect that can be attached to any agent.
• Council mode – on‑demand multi‑model consensus (@council).
• Companion UI – optional floating window that shows active agents.
• Multiplexer integration – live view of each agent in a separate pane of Tmux, Zellij, Kitty, WezTerm, etc.
• Preset switching – change the whole team’s models at runtime with /preset; presets can inherit from one another.
• Extensive code‑intelligence tools – LSP‑style helpers, AST‑aware search across ~25 languages, built‑in MCPs for docs and GitHub search.
• Full configurability – per‑agent model, skills, MCP whitelist/blacklist, global disable switches, and project‑local overrides.
• Marketplace – community‑contributed agent packages that can be installed without restarting OpenCode.

Quick installation

# Using Bun (preferred)
bunx oh-my-opencode-slim@latest install
# Or with npm/npx if you don’t have Bun
npx oh-my-opencode-slim@latest install

The installer writes the plugin path into ~/.config/opencode/opencode.json so OpenCode loads it automatically.

For a development copy you can clone the repo, run bun install && bun run build, then execute bun dist/cli/index.js install.

Getting started after install

1. Log in to the LLM providers you plan to use: opencode auth login.
2. Refresh the model list: opencode models --refresh.
3. Edit the generated config at ~/.config/opencode/oh-my-opencode-slim.json – choose models for each agent or keep the defaults.
4. (Optional) Enable Multiplexer integration to watch each agent in its own terminal pane.
5. Verify everything is alive: opencode → ping all agents.

Typical use‑case flow

User: "Add a new feature to the payment module that logs every transaction."
→ Orchestrator decides the work needs code‑base scouting, design, implementation, and verification.
→ Explorer maps the payment module files.
→ Designer drafts a UI/CLI sketch for the log viewer.
→ Librarian pulls the latest logging library docs.
→ Fixer writes the implementation.
→ Oracle reviews the overall architecture.
→ Orchestrator reconciles all outputs and returns a single, polished answer.

The user can also manually invoke a specialist (@explorer list all routes) or trigger a Council (@council which logging library is safest).

Who would benefit?

• Developers who use OpenCode and want richer, multi‑model assistance without manually switching providers.
• Teams that need a reproducible “AI workflow” where cost‑sensitive tasks stay cheap while critical design decisions get the best model.
• Tool builders who want to extend OpenCode with custom agents via the marketplace or their own configuration.

---

TL;DR

oh‑my‑opencode‑slim is a real, installable plugin that turns OpenCode into a small AI agency. It lets you mix‑and‑match models, run many agents in parallel, and customize the whole pipeline through a JSON config and a set of ready‑made “skills”. If you already work with OpenCode, installing the plugin gives you a more modular, faster, and potentially cheaper AI‑coding assistant.