OpenCode Tutorial

OpenCode is an open-source AI coding assistant that runs in your terminal (TUI-based). It follows the concept of Vibe Coding, where the user acts as a commander rather than an executor. Instead of manually finding files and editing code, you provide natural language instructions, and the AI understands your project, modifies code, executes commands, and completes multi-step tasks automatically.

Key advantages of OpenCode over IDE-based tools (like Cursor or GitHub Copilot) include:

• Model Freedom: Supports over 75 models (Claude, GPT, Gemini, Llama, DeepSeek, Qwen, etc.).
• Open Source: Provides control and data privacy via direct API or local model usage.
• TUI-based Workflow: Lightweight, fast, and suitable for automation or CI/CD embedding.

The Advanced Customization stage (Phase 5) is designed as a reference manual for developers looking to deeply customize OpenCode. Instead of a linear tutorial, it is intended to be used as a lookup guide when specific needs arise.

By exploring these advanced modules, you can master the complete configuration system, create custom Agents, Skills, and commands, configure permissions and themes, implement MCP extensions, and integrate OpenCode with IDEs or remote environments.

The OpencodeClient class provides access to 21 specialized API modules. These modules allow developers to manage sessions, files, projects, configuration, authentication, and more. Key modules include:

• session: Core module for message sending and history management.
• project: Project management.
• file: File operations.
• find: Search functionality.
• config: Configuration management.
• auth: Authentication management.
• tui: TUI (Terminal User Interface) control.
• mcp: MCP server management.
• lsp: LSP server status.

Phase 2 focuses on mastering the OpenCode interface and core operations required for daily AI coding tasks. The primary goal is to enable users to perform file operations, code reading, and text generation independently.

Key Learning Objectives:

• Familiarity with the OpenCode interface and core operations.
• Mastery of session management, including creating, switching, undoing, and exporting sessions.
• Proficiency with core command triggers: @ for referencing files, ! for executing commands, and / for slash commands.

OpenCode provides several key capabilities for working with AI assistants:

• Global Model Support: Native support for various models including Claude, OpenAI, DeepSeek, and Ollama.
• Multi-file Context: Ability to read entire projects at once to understand code relationships, rather than feeding files one by one.
• Deep Customization: Support for creating custom Agents, Skills, Commands, and Themes to build personalized AI tools.
• Source Code Analysis: Features are taught based on official OpenCode source code analysis rather than theoretical tutorials.

The OpenCode Quick Reference Guide (速查手册) provides a centralized directory for all technical specifications and troubleshooting resources. Use this guide to navigate between keyboard shortcuts, slash commands, CLI references, configuration options, model providers, and prompt templates.

How to use the guide:

• Troubleshooting: Consult FAQ for common questions or Troubleshooting for specific error solutions.
• Finding Commands: Refer to Slash Commands for interactive commands or CLI Reference for automation scripts.
• Selecting Models: Check Model Providers to see the 75+ supported models.
• Prompting: Browse the Prompt Template Library for inspiration or follow Prompt Engineering Guidelines for high-quality template creation.

Chrome DevTools MCP is an official Google browser MCP that allows an AI agent to connect directly to your active Chrome browser session.

Unlike traditional browser automation (like Puppeteer or Playwright) which typically launches a fresh, isolated instance, Chrome DevTools MCP:

• Reuses your existing logged-in state: You don't need to sign in to websites again.
• Connects to an already open Chrome: It works with your current session.
• Uses uid for element targeting: Instead of CSS or XPath, it uses temporary identifiers from the page's accessibility tree. Note that uid values are not permanent and will change if the page refreshes.
• Provides DevTools-level access: The AI can read the Network panel, inspect the Elements panel, and analyze performance data directly.

The Agent Client Protocol (ACP) is an open protocol that standardizes communication between code editors and AI coding agents.

OpenCode implements this protocol by running opencode acp as a child process. The editor communicates with the process via stdin/stdout using the nd-JSON (newline-delimited JSON) format for JSON-RPC communication.

Communication Flow: Editor <-> JSON-RPC (stdio) <-> opencode acp

A rule file is a Markdown file containing instructions you want the AI to follow. OpenCode automatically reads these files and injects their content into the system prompt for every message sent.

Key Feature: Hot-Reloading Rule file changes take effect immediately. You do not need to restart OpenCode or create a new session after modifying a rule file. OpenCode re-reads the file every time a message is sent.