What Is OpenCode? The Open-Source AI Coding Agent That Just Hit #1 on Hacker News

Why AI Coders Need This

Here's the problem with most AI coding agents right now: they lock you in.

Claude Code is excellent — arguably the best AI coding agent available today. But it only works with Anthropic's Claude models. If Anthropic raises prices, changes their terms, or a better model comes along from Google or OpenAI, you're stuck. Your workflows, your muscle memory, your project configurations — all tied to one vendor.

Codex from OpenAI has the same issue in reverse. Great tool, locked to OpenAI's ecosystem.

Think of it like construction. Imagine if your favorite power drill only worked with one brand of screws. And every time you wanted to use a different type of screw — even if it was cheaper, stronger, or better for the job — you'd need to buy an entirely new drill. That's what vendor lock-in feels like in AI coding tools.

OpenCode is the universal drill. It's a powerful AI coding agent that works with any AI model from any provider. Claude, GPT, Gemini, open-source models running on your own machine — all through the same interface. Same commands. Same workflow. Same muscle memory.

And because it's open source, no company can pull the rug out from under you. The code is public. The community maintains it. If the original team disappeared tomorrow, the tool would keep going.

For vibe coders — people building real software with AI as their primary coding partner — this matters. You're not just picking a tool for today. You're picking the foundation for how you'll work with AI for the next several years. OpenCode is the first serious option that doesn't make you bet everything on one company.

What It Does

OpenCode is an AI coding agent. That means it doesn't just suggest code in a sidebar — it actively reads your codebase, writes files, runs commands, and makes changes on your behalf. You have a conversation with it, and it does the work.

Here's what that looks like in practice:

Terminal Interface (TUI)

The primary way to use OpenCode is through its terminal interface — a beautiful text-based UI built by the creators of terminal.shop (the same team behind the popular Bubble Tea framework). You open your terminal, navigate to your project folder, type opencode, and start talking to it.

It looks like a chat app inside your terminal. You type what you want. The AI responds, shows you what it plans to do, and makes the changes. All without leaving the command line.

Desktop App & IDE Extension

Not a terminal person? OpenCode also ships as a desktop application (macOS, Windows, Linux) and an IDE extension. Same AI agent, different wrappers. The desktop app gives you a standalone window. The IDE extension puts it right inside your editor.

Any AI Model, Any Provider

This is the big differentiator. OpenCode supports 75+ AI providers through Models.dev:

• Anthropic Claude — Claude Opus, Sonnet, Haiku
• OpenAI GPT — GPT-4o, GPT-5, and newer models
• Google Gemini — Gemini 2.5 Pro, Flash, and more
• GitHub Copilot — log in with your GitHub account to use your existing Copilot subscription
• ChatGPT Plus/Pro — log in with your OpenAI account to use your ChatGPT subscription
• AWS Bedrock, Azure OpenAI, Groq, OpenRouter — enterprise and alternative providers
• Local models — run open-source models on your own machine with zero cloud dependency

You can switch between models mid-project. Start with a fast, cheap model for simple tasks. Switch to Claude Opus or GPT-5 for complex architecture decisions. Use a local model when you're working on sensitive code that shouldn't leave your machine. All through the same tool.

Built-in LSP Support

Here's something technical that has a very practical impact: OpenCode includes Language Server Protocol (LSP) support out of the box. In plain English, this means OpenCode automatically understands the structure of your code — where functions are defined, what types variables have, which files import which other files. It loads the right language intelligence for whatever language your project uses.

Why does this matter to you? Because the AI makes fewer mistakes. When OpenCode knows the structure of your code (not just the text), it produces more accurate edits and catches issues that a text-only AI would miss.

Plan Mode and Build Mode

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

• Build mode (default) — Full access. The AI can read files, write code, run commands, and make changes to your project.
• Plan mode — Read-only. The AI analyzes your codebase and suggests what to do without actually changing anything. It asks permission before running commands.

This is huge for vibe coders. You can ask OpenCode to plan a feature first — "How would you add user authentication to this app?" — review the plan, give feedback, and only then switch to Build mode to execute. It's like asking a contractor for a detailed estimate before they start swinging hammers.

Multi-Session Support

You can run multiple OpenCode agents in parallel on the same project. One agent works on the frontend. Another handles the API. A third writes tests. Each operates in its own session with its own conversation history.

Session Sharing

Built a feature and want to show your team what the AI did? OpenCode lets you share a link to any session. Your collaborator (or future you) can see the entire conversation — what you asked, what the AI suggested, what code was written. It's like handing someone a recording of the entire build process.

Auto-Compact for Long Sessions

Every AI coding tool struggles with long conversations — the AI eventually "forgets" what you discussed earlier as the context window fills up. OpenCode handles this automatically. When your conversation approaches the model's context limit, it summarizes everything and starts a fresh session with the summary, so you can keep working without losing context.

Undo/Redo

Asked the AI to refactor something and it made a mess? Type /undo and all changes are reverted. Your original prompt reappears so you can try again with better instructions. Changed your mind? /redo brings the changes back. Simple, powerful, essential.

How It Compares

The AI coding agent landscape is moving fast. Here's how OpenCode stacks up against the tools vibe coders are actually using:

OpenCode vs. Claude Code — The Direct Comparison

This is the comparison everyone's making, and OpenCode's own documentation addresses it directly. Here's the honest breakdown:

Claude Code produces better output — when using Claude models. That's because Anthropic has deeply optimized Claude Code for their own models. The prompting, the context management, the tool use — it's all tuned for Claude. When you use Claude through Claude Code, you get the absolute best that Claude can deliver.

OpenCode gives you freedom. It's 100% open source. It works with any provider. It has features Claude Code doesn't — LSP support for smarter code intelligence, multi-session for parallel work, a client/server architecture that lets you run it on your computer and control it remotely from a mobile app. And if a new model comes out tomorrow that's better than Claude, you can switch to it instantly without changing tools.

The practical difference: If you're using Claude models and money isn't a concern, Claude Code gives you the best experience today. If you want flexibility, if you use multiple AI providers, if you care about open source, or if you're budget-conscious and want to use cheaper models for routine work — OpenCode is the better choice.

OpenCode vs. Codex

Codex is OpenAI's AI coding agent. It runs in the cloud with sandboxed execution — meaning it spins up a virtual environment to run your code safely. That's great for automated tasks and CI/CD pipelines, but it means Codex doesn't have the same direct access to your local files and tools that OpenCode does.

OpenCode runs on your machine, in your terminal, with direct access to your project files, your git history, your local development tools. It's more hands-on. Codex is more hands-off — you give it a task and it works in the cloud. Different philosophies for different workflows.

OpenCode vs. Cursor

Cursor is an AI-powered IDE — a visual editor (forked from VS Code) with AI deeply integrated into the editing experience. OpenCode is a terminal-first agent. The difference is philosophical: Cursor gives you a visual workspace where AI assists. OpenCode gives you a conversation where AI acts.

Many vibe coders use both. OpenCode for major feature builds, refactoring, and complex multi-file changes. Cursor for day-to-day editing, debugging, and visual code review. They complement each other rather than compete.

When to Use It

OpenCode excels in specific situations. Here's where it genuinely shines:

• You want model freedom: You're using Claude today but curious about Gemini 2.5 Pro. Or you want to use a cheap model for simple tasks and a premium model for complex ones. OpenCode lets you switch providers without switching tools.
• You're budget-conscious: Claude Code requires an Anthropic API subscription. OpenCode is free — you can use GitHub Copilot (included with many developer plans), local models (completely free), or pay-as-you-go through OpenCode Zen ($20 balance, no subscription).
• You care about privacy: OpenCode doesn't store your code or context data. You can run it with local models and keep everything on your machine. No cloud, no data sharing, no third-party access to your codebase.
• You want open source: The code is on GitHub. You can inspect it, modify it, contribute to it. No black box. No surprise changes to terms of service. The community has over 800 contributors keeping it moving.
• You run multiple projects: Multi-session support means you can have several agents working on different parts of your project simultaneously. One handles the API, another the frontend, a third writes tests — all in parallel.
• You want a terminal-first workflow: If you're comfortable in the terminal (or learning to be), OpenCode's TUI is one of the best-designed terminal interfaces available. It was built by the team behind Bubble Tea and terminal.shop — people who live in the terminal.

When NOT to Use It

Honesty time. OpenCode isn't the right tool for everyone or every situation:

• You've never used a terminal: OpenCode's primary interface is a terminal UI. If you've never opened a terminal before, the learning curve is real. Start with Cursor (visual, familiar, lower barrier) and come back to OpenCode when you're comfortable with the command line.
• You want the absolute best Claude experience: If you're committed to Claude models and want the highest possible quality, Claude Code is still the better choice. Anthropic has optimized it specifically for their models in ways an open-source tool can't match.
• You need a full visual IDE: OpenCode is an agent, not an editor. It doesn't have a file tree, a visual diff viewer, or inline code highlighting the way Cursor does. If you need those visual tools, Cursor or VS Code + Copilot is a better fit.
• You need hand-holding through setup: OpenCode requires you to install it (via npm, Homebrew, or the install script), configure an AI provider, and navigate to your project in the terminal. It's not complicated, but it's not "click a button and start talking" either. Tools like Cursor have a gentler onboarding.
• Your team is standardized on one provider: If your team already uses Claude Code or Codex and everyone's workflow is built around that tool, switching to OpenCode introduces friction. The flexibility benefit matters less when your team has already committed to a provider.

What AI Gets Wrong About OpenCode

If you ask other AI tools about OpenCode, you'll get some inaccurate or outdated information. Here's what to watch for:

"OpenCode Is Just a Claude Code Clone"

It's not. While the core concept is similar (terminal-based AI coding agent), OpenCode has features Claude Code doesn't: multi-provider support, built-in LSP integration, multi-session parallel agents, session sharing, a client/server architecture, and a desktop app. OpenCode was built from scratch by the team at Anomaly — the same people behind terminal.shop and the SST framework. It's its own product with its own vision.

"Open Source Means Lower Quality"

This is a common misconception. With 120,000+ GitHub stars and 800+ contributors, OpenCode has more developer attention than most commercial tools. The quality gap isn't in the tool itself — it's in model-specific optimization. Claude Code is optimized for Claude. OpenCode is optimized for flexibility. When you use Claude models through OpenCode, the output is very close to Claude Code. When you use other models, OpenCode often performs better than those models' native tools because of its LSP integration and context management.

"You Need to Be a Developer to Use It"

You need terminal basics — opening a terminal, navigating folders, running commands. That's a 30-minute learning curve, not a computer science degree. If you can follow a recipe, you can use OpenCode. Our complete guide to vibe coding covers the terminal fundamentals you need.

"It's Free So There Must Be a Catch"

The tool is free. The AI models cost money (or can be free if you use local models or included services like GitHub Copilot). OpenCode makes money through Zen — their curated model service at $20 pay-as-you-go with zero markup. They're also building enterprise features. But the core tool is genuinely free, genuinely open source, and you're never required to use their paid services.

"It's Too New to Trust"

5 million developers use it monthly. 120,000+ GitHub stars. 800+ contributors. 10,000+ commits. These aren't early-stage numbers. OpenCode is production-grade and battle-tested, even if it's newer to the mainstream spotlight.

The OpenCode Workflow for Vibe Coders

Here's the step-by-step workflow that gets the most out of OpenCode if you're a vibe coder:

1. Install OpenCode: The fastest way is the one-line install script:
   curl -fsSL https://opencode.ai/install | bash
   Or install via npm (npm install -g opencode-ai), Homebrew (brew install anomalyco/tap/opencode), or download the desktop app from opencode.ai/download.
2. Connect an AI provider: Run opencode in your terminal, then type /connect. Choose your provider — OpenCode Zen for the easiest setup ($20 pay-as-you-go), or connect your own API keys from Anthropic, OpenAI, Google, GitHub Copilot, or others.
3. Initialize your project: Navigate to your project folder and run opencode. Then type /init. OpenCode analyzes your project structure and creates an AGENTS.md file that helps it understand your codebase — like giving a new team member the project onboarding doc.
4. Plan before you build: Hit Tab to switch to Plan mode. Describe the feature you want. OpenCode will analyze your code and suggest an implementation plan without making any changes. Review the plan. Give feedback. When it looks right, Tab back to Build mode and say "Go ahead."
5. Iterate with context: As you work, OpenCode remembers the full conversation. Reference earlier decisions: "Remember the auth pattern we discussed? Apply that to the settings route too." The AI maintains context until it approaches the model's limit, then auto-compacts to keep going.
6. Share and collaborate: Built something cool? Type /share to create a shareable link to the entire conversation — great for showing teammates what the AI did, getting code review, or debugging together.

What to Learn Next

OpenCode is one piece of the AI coding tool landscape. Understanding how the pieces fit together makes you a more effective builder:

• Claude Code Beginner's Guide — The premium proprietary alternative. Understand what you're comparing against and when it's worth the cost.
• Codex vs. Claude Code — The two biggest proprietary AI coding agents compared. Useful context for understanding where OpenCode fits as the open-source third option.
• Cursor Beginner's Guide — The visual AI IDE. Many vibe coders use Cursor alongside a terminal agent like OpenCode for different types of work.
• How to Debug AI-Generated Code — Essential reading for when any AI tool (including OpenCode) produces code that doesn't work. The debugging strategies apply universally.
• The Complete Guide to Vibe Coding — The full landscape of AI-enabled development. How all these tools connect, and how to build a workflow that works for you.
• Local AI Models for Coding — OpenCode supports local models through Ollama and LM Studio. If you want full privacy and zero API costs, this guide covers the setup.
• What Is Agentic Coding? — OpenCode is an agentic coding tool. Understanding the agentic paradigm helps you use it (and its competitors) more effectively.

FAQ