OpenCode Open source

What it is

OpenCode is an open-source (MIT) AI coding agent built around a polished terminal interface, developed in the open at anomalyco/opencode. It is provider-neutral: you connect API keys from the LLM providers you already use, including local models, or use OpenCode Zen, the project's own gateway of tested models. A desktop app is available in beta for macOS, Windows and Linux.

The README's install script is curl -fsSL https://opencode.ai/install | bash. The docs also list npm install -g opencode-ai, brew install anomalyco/tap/opencode, Chocolatey, Scoop and Docker.

Key features

• Two built-in agents you switch with Tab: build (full access) and plan (read-only analysis)
• A general subagent for searches and multi-step work
• /init scans your project and writes an AGENTS.md file
• Language server (LSP) integration for code-aware edits
• Session sharing links and GitHub/GitLab integration
• Works with many providers, so you can bring an existing subscription or key

Skills, subagents and MCP

OpenCode is one of the closest open-source matches to Claude Code in extensibility. Custom agents are Markdown files in .opencode/agents/ (project) or ~/.config/opencode/agents/ (global), or entries in opencode.json; each can be a primary agent or a subagent you call with @, with its own model and tool permissions. The docs also cover Agent Skills, MCP servers and plugins. Project instructions go in AGENTS.md.

Pricing

OpenCode is free to use. You pay whichever model provider you connect. OpenCode Zen is optional and pay-as-you-go per token, with some free models on the list and a team spending cap; at the time of writing (September 2026) rates vary widely by model, so check the Zen docs for current prices.

Who it is for

Developers who want Claude Code-style agents, subagents and skills without being tied to one model vendor, and anyone who wants to run local or low-cost models.

Verdict

OpenCode combines a fast, well-designed terminal UI with a flexible configuration model and no licence cost. The trade-off is that you manage model costs and quality yourself, and the ecosystem of ready-made extensions written specifically for it is smaller than Claude Code's, although skills in the shared SKILL.md format carry over.