OpenCode: The Open Source AI Coding Agent with 100K Stars

AI Editors & App Builders
1

opencode-wordmark-simple-dark
opencode-wordmark-simple-dark1282×230 1.11 KB

If you’ve been using Claude Code, Cursor, or GitHub Copilot and wondering whether the open source world had a serious answer — it does. OpenCode is a fully open source, provider-agnostic AI coding agent that has exploded to 100K+ GitHub stars, 700+ contributors, and 2.5 million monthly developers. And it’s free.

What is OpenCode?

OpenCode is a terminal-first AI coding agent built in Go by the team at Anomaly — the same folks behind terminal.shop. It’s available as a CLI tool, a desktop app (macOS, Windows, Linux beta), and an IDE extension for VS Code, Cursor, JetBrains, Zed, Neovim, and Emacs.

Think of it as Claude Code’s open source cousin — but with one crucial difference: it works with any model from any provider. Claude, GPT, Gemini, Groq, local models through LM Studio — you name it. OpenCode supports 75+ LLM providers through Models.dev, their open source model database.

Even better: you can use your existing ChatGPT Plus/Pro subscription or your GitHub Copilot account to power it. No extra API keys required.

The Terminal UI Is Gorgeous

Built by self-described neovim users using Bubble Tea, the TUI is arguably the best-looking terminal interface in any AI coding tool. Tool calls are clearly formatted, file diffs are readable, and the whole experience feels polished — not like an afterthought bolted onto a CLI.

Dual Agent System

OpenCode ships with two built-in agents you switch between with the Tab key:

  • Build — Full-access agent for development work. It can read, write, execute, and modify your codebase.
  • Plan — Read-only mode for analysis and code exploration. It refuses to modify files and asks permission before running commands. Perfect for exploring unfamiliar codebases or reviewing before committing to changes.

There’s also a @general subagent for complex searches and multistep tasks that runs in parallel.

You can even create custom agents using markdown files with specific system prompts, tool permissions, and model configurations. A review.md file becomes a review agent. Simple.

LSP Integration — The Secret Weapon

This is where OpenCode quietly stands apart. It automatically loads Language Server Protocol servers for 40+ languages — Rust, TypeScript, Python, Swift, Terraform, and more. This means the LLM gets real compiler feedback, not just text pattern matching. The difference in code quality is noticeable.

MCP Support

OpenCode supports both local (STDIO) and remote (HTTP/SSE) MCP servers. Connect it to databases, APIs, GitHub — whatever tools your workflow needs. One caveat from the docs: some MCP servers (particularly GitHub’s) can be token-hungry, so watch your context usage.

Client-Server Architecture

Here’s a forward-thinking design decision: OpenCode uses a client-server architecture. The agent runs on your machine, but you can drive it remotely — even from a mobile app. The TUI is just one possible frontend. This opens up interesting possibilities for remote development workflows.

Privacy First

OpenCode does not store your code or context data. Period. This makes it viable for privacy-sensitive environments and enterprise use cases where tools like Cursor (which uploads code to servers) are non-starters.

What Does It Cost?

OpenCode itself is completely free and open source (MIT license). You pay only for the LLM usage from whatever provider you choose. They also offer OpenCode Zen — a curated set of models they’ve benchmarked specifically for coding agents, so you don’t have to guess which model performs best.

For enterprise needs, they offer additional features around team management and compliance.

How It Compares

vs Claude Code: Claude Code is more mature and polished, but locks you into Anthropic’s ecosystem and pricing. OpenCode gives you provider freedom — if a better model drops next month or your budget changes, just switch.

vs Cursor: Cursor is a full IDE (VS Code fork) at $20-200/month. OpenCode is terminal-first with IDE extensions, and free. Different philosophies — IDE-centric vs terminal-centric.

vs GitHub Copilot: Copilot is primarily autocomplete + chat with file-level context. OpenCode is a full autonomous agent with project-wide context through LSP.

Quick Start

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

# Or via package managers
npm i -g opencode-ai@latest
brew install anomalyco/tap/opencode

Why It Matters for LatAm Developers

The free + pay-only-for-LLM-usage model is huge for developers in our region where $20-200/month IDE subscriptions add up fast. Being able to use local models means you can experiment with zero cost. And the provider flexibility means you can hunt for the best price-to-performance ratio across providers — use a cheap model for planning, switch to Claude or GPT for execution.

Links