Skip to main content

​
Install Oh My OpenCode

The fastest way to get started is to let an AI agent handle the installation:
1

Paste this into your LLM agent

Copy this prompt to Claude Code, AmpCode, Cursor, or any LLM agent:
Install and configure oh-my-opencode by following the instructions here:
https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
Seriously, let an agent do it. Humans fat-finger configs. The agent will ask about your subscriptions and configure everything automatically.
2

Run the interactive installer (alternative)

If you prefer manual installation, run:
bunx oh-my-opencode install
The CLI ships with standalone binaries for all major platforms:
  • macOS (ARM64, x64)
  • Linux (x64, ARM64, Alpine/musl)
  • Windows (x64)
No runtime (Bun/Node.js) required after installation.
3

Configure your subscriptions

The installer will ask about your AI subscriptions:
  • Claude Pro/Max — Recommended for Sisyphus (main agent)
  • ChatGPT Plus — Enables GPT-5.2 for Oracle and GPT-5.3-codex for Hephaestus
  • Gemini — Great for visual/frontend tasks
  • GitHub Copilot — Fallback provider when native providers unavailable
Sisyphus strongly recommends Claude Opus 4.6. Using other models may result in significantly degraded experience.
4

Authenticate providers

After installation, authenticate your chosen providers:
opencode auth login
Select your provider (Anthropic, Google, GitHub) and follow the OAuth flow.

​
Your first ultrawork command

Once installed, navigate to any code project and run: 
opencode
Then type: 
ultrawork: Add a health check endpoint to the API
That’s it. Watch what happens:
1

Intent classification

The Intent Gate analyzes your request to understand the true goal — not just literal words.
2

Codebase exploration

Sisyphus spawns Explore agent to grep the codebase for existing API patterns, routes, and health check conventions.
3

Implementation

Based on discovered patterns, Sisyphus delegates to the right category agent:
  • quick category for simple endpoint addition
  • ultrabrain if complex business logic required
4

Verification

LSP diagnostics run automatically. Type errors? Fixed. Import issues? Resolved. Build errors? Addressed.
5

Completion

Todo enforcer ensures the agent doesn’t stop until 100% complete. No halfway implementations.
You can type just ulw instead of ultrawork. It’s an alias for the lazy (like all of us).

​
What just happened?

Unlike single-model tools that process one thing at a time, Oh My OpenCode ran multiple agents in parallel: This is parallel execution. While one agent writes code, another researches patterns, another checks documentation. Like having 5 engineers instead of 1.

​
Two working modes

​
Ultrawork mode: For the lazy

Just include ultrawork (or ulw) in your prompt: 
ultrawork: Refactor the auth module to use JWT tokens
The agent figures everything out. Explores, researches, implements, verifies. Keeps working until done. This is the “just do it” mode. Full automatic.

​
Prometheus mode: For the precise

Press Tab to enter Prometheus mode for complex multi-day tasks:
1

Interview phase

Prometheus asks clarifying questions like a real engineer:
  • What’s the current authentication mechanism?
  • Should we support token refresh?
  • What’s the expected token expiration time?
  • Do we need to migrate existing sessions?
2

Plan creation

Prometheus builds a detailed plan with:
  • Task breakdown with dependencies
  • Verification criteria for each step
  • Edge cases and error scenarios
  • Rollback strategy if needed
3

Execution

Run /start-work to activate Atlas:
  • Distributes tasks to specialized subagents
  • Tracks progress across sessions
  • Accumulates learnings (conventions discovered in task 1 passed to task 5)
  • Verifies each completion independently
Use Prometheus for:
  • Multi-day projects
  • Critical production changes
  • Complex refactoring
  • When you want a documented decision trail

​
Common use cases

ultrawork: Fix all TypeScript compilation errors
Sisyphus will:
  1. Run lsp_diagnostics to get all errors
  2. Categorize errors by type (import issues, type mismatches, etc.)
  3. Delegate fixes to category agents in parallel
  4. Re-run diagnostics after each batch
  5. Continue until zero errors remain
ultrawork: Add user profile editing with avatar upload
The agent will:
  1. Research existing user management patterns
  2. Design database schema changes if needed
  3. Implement backend API endpoints
  4. Create frontend UI components
  5. Add validation and error handling
  6. Write tests
  7. Update documentation
Press Tab for Prometheus mode:
Refactor the payment processing module to support multiple payment providers
Prometheus will interview you about:
  • Which providers to support initially
  • How to handle provider-specific features
  • Migration strategy for existing payments
  • Testing approach for each provider
Then /start-work executes with full orchestration.
@oracle Explain the authentication flow in this codebase
Oracle (read-only consultant) will:
  1. Find auth-related files using LSP and grep
  2. Trace the flow from login to session management
  3. Identify security patterns and potential concerns
  4. Explain with code references (file:line format)

​
Next steps

Complete installation guide

Set up all providers, configure agents, and understand model matching

Learn about agents

Deep dive into Sisyphus, Hephaestus, Prometheus, and the supporting cast

Understand orchestration

How agents collaborate, delegate, and accumulate learnings

Customize configuration

Override models, disable features, and tune for your workflow

Build docs developers (and LLMs) love

Get started for freeTalk to us