OpenCode: The 95K-Star AI Coding Agent Built for the Terminal

“The open source AI coding agent. Free models included or connect any model from any provider.” — opencode.ai

The Numbers

While OpenClaw dominated headlines and Moltbook made international news, OpenCode quietly became the most-used open-source AI coding agent for the terminal.

What OpenCode Does

OpenCode is an open-source coding agent that runs in the terminal, desktop app, or IDE. It connects to any LLM provider and helps developers write, edit, and debug code through natural language interaction.

Key Differentiators

• Model agnostic: Works with Claude, GPT, Gemini, DeepSeek, local models via Ollama, and 75+ providers through Models.dev. Switch providers without changing your workflow.
• LSP-enabled: Automatically loads the correct Language Server Protocol for the language the LLM is working with. The agent understands type systems, navigates definitions, and catches errors like an IDE.
• Multi-session: Start multiple agents in parallel on the same project. Run one on the frontend while another handles the backend.
• Existing subscription support: Log in with your GitHub Copilot account, or your ChatGPT Plus/Pro subscription. No new subscription required.
• Share links: Generate a shareable link to any session for debugging or reference.
• Privacy-first: Does not store code or context data. Runs locally, communicates with AI providers only when making API calls.

Installation

OpenCode supports five installation methods:

# curl
curl -fsSL https://opencode.ai/install | bash

# npm
npm install -g opencode

# bun
bun install -g opencode

# brew (macOS)
brew install opencode

# paru (Arch Linux)
paru -S opencode

A desktop app is also available in beta for macOS, Windows, and Linux.

OpenCode vs. Claude Code

The two most commonly compared terminal coding agents serve different developer profiles:

The tradeoff is clear: OpenCode gives you model flexibility and simplicity. Claude Code gives you ecosystem depth and workflow automation. The right choice depends on whether you need a flexible coding assistant or a complete development workflow.

The Zen Tier

OpenCode offers a “Zen” tier that provides access to a handpicked set of AI models the team has tested and benchmarked specifically for coding agents. This addresses a real pain point: not all models perform equally well for agentic coding tasks, and developers waste time discovering this through trial and error.

Zen provides validated models that work, so developers can focus on building rather than debugging provider inconsistencies.

Why 2.5 Million Developers Chose It

OpenCode's adoption reflects a clear developer preference:

• No lock-in: If Anthropic raises prices, switch to GPT. If OpenAI is slow, switch to Gemini. The agent stays the same.
• Use what you have: GitHub Copilot and ChatGPT Plus are already common subscriptions. OpenCode lets you use them for agentic coding without paying for another service.
• Works immediately: One install command, one config step, and the agent is productive. No hooks to configure, no plugins to install, no memory files to maintain.
• Open source trust: Developers can inspect the code, verify privacy claims, and contribute improvements.

Where It Falls Short

OpenCode intentionally trades ecosystem depth for simplicity:

• No workflow automation: Without hooks, developers cannot automate pre-commit checks, quality gates, or session-level learning.
• Limited memory: No equivalent to Claude Code's CLAUDE.md system for maintaining project context across sessions.
• No MCP server ecosystem: Cannot natively connect to databases, GitHub APIs, or documentation services the way Claude Code can through MCP.
• Simpler subagent model: Claude Code's specialized subagents (explore, plan, build, review) have no direct equivalent.

Implications for DevRel

OpenCode's success demonstrates that developer tool adoption follows a predictable hierarchy:

1. Does it work out of the box? Friction at install time kills adoption faster than missing features.
2. Does it respect existing investments? Supporting GitHub Copilot and ChatGPT subscriptions is a distribution strategy, not just a feature.
3. Is the source available? For developer tools that touch code, open source isn't optional — it's table stakes for trust.
4. Can I switch away? Model-agnostic design reduces perceived risk and encourages experimentation.

The best coding agent is the one that stays out of your way. OpenCode doesn't try to be a platform. It opens in your terminal, connects to your model, and helps you write code. 2.5 million developers chose that over the noise.