OpenCode Overview

This section provides a cleaner documentation map for day-to-day use. It is based on a local setup that uses OpenCode, CLIProxyAPI, and oh-my-openagent, with optional extras such as Superpowers and Impeccable.

Instead of one long guide, the same information is organized here by task: install, configure, extend, operate, and troubleshoot.

The setup described here does not assume CLIProxyAPI is the only provider path. OpenCode can also work directly with providers such as opencode and opencode-go. In this repository, cliproxyapi is mainly useful when you want to place several Codex accounts behind a single OpenAI-compatible endpoint and let the proxy handle routing or load balancing.

Documentation Map

• Quick Start for the full stack
• Install OpenCode and CLIProxyAPI
• Configure providers, agents, and models
• Choose between opencode, opencode-go, and cliproxyapi
• Set up workflows, MCP tools, and project rules
• Configure MCP tools
• Automate project workflow with AGENTS.md
• Create custom local skills
• Use planner strategy to save tokens
• Reference: OpenCode agent modes
• Tutorial: Image to code workflow
• Use RTK to reduce token usage
• Reference: Superpowers skills catalog
• Guide: When to use Impeccable
• Launch OpenCode and use model variants
• Troubleshoot common OpenCode issues

Architecture at a Glance

┌─────────────────────┐
│      OpenCode       │
│  oh-my-openagent    │
└─────────┬───────────┘
          │ OpenAI-compatible API
          │ http://localhost:8317/v1
          ▼
┌─────────────────────┐
│     CLIProxyAPI     │
│ Homebrew service    │
│ Port 8317           │
└─────────┬───────────┘
          │ OAuth tokens / provider auth
          ▼
┌─────────────────────┐
│ Upstream LLM models │
│ GPT-5.x, Claude,    │
│ Gemini, Kimi, etc.  │
└─────────────────────┘

What This Setup Optimizes For

• One local endpoint for many models through CLIProxyAPI
• OpenCode as the main coding interface
• oh-my-openagent for agent orchestration
• ECC for specialized agents (64 agents, 262 skills, 84 commands) — see ECC guide
• Caveman for token compression (~75% fewer output tokens) — see Caveman guide

Complementary Tools

For advanced workflows, consider adding:

• ECC — Agent harness with planners, reviewers, security scanners, and continuous learning. Works across Claude Code, OpenCode, Codex, and more.
• Caveman — Token compression that cuts ~75% output tokens with no accuracy loss. One-line install, works across 30+ agents.

Naming note

The upstream project is now called oh-my-openagent, but many current plugin and config examples still use legacy names such as oh-my-opencode, oh-my-opencode.json, and plugin: ["oh-my-opencode"]. This docs section uses the current project name for explanation, while preserving the real config names where they still matter.

• Optional MCP tools like Context7, Playwright, and React Grab
• Project-level workflow rules with AGENTS.md

Recommended Reading Order

1. Start with Quick Start for the full stack
2. Continue to Install OpenCode and CLIProxyAPI
3. Decide your provider path in Choose between opencode, opencode-go, and cliproxyapi
4. Configure providers, agents, and models in Configuration
5. Add workflow and tool layers with MCP Guide, Project Rules, and Skills Integration
6. Learn efficiency patterns in Use planner strategy to save tokens and Reference: OpenCode agent modes
7. Learn the runtime workflow in Launch OpenCode and use model variants
8. Use Troubleshooting when something breaks

Notes

• This guide is written for the current macOS-based setup in this repository.
• It intentionally splits the setup into smaller task-based pages instead of one giant document.
• The full OpenCode setup now lives in this section so it can be maintained in one place.