OpenCode calls itself an open source AI coding agent. You can use it as a terminal interface, as a desktop app, or as an IDE extension, and the code is published under the MIT license. It has an active GitHub repository and is installed through the usual routes: an install script, npm, Bun, pnpm, Yarn, Homebrew, and several Linux and Windows package managers, plus Docker.
The workflow is built around two built-in agents. Build has full access to make changes; plan is read-only, so you can have it analyze the code and propose an implementation before anything is touched. Around that sit practical features: undo and redo for modifications, a /share command that produces a link to a conversation, fuzzy file search with the @ key, image drag-and-drop into prompts, and an AGENTS.md file that records what it learns about your project. It also supports LSP servers and MCP servers, which is how it reaches language tooling and outside systems.
On models, the docs say it works with any LLM provider through an API key. For newcomers they recommend OpenCode Zen, a curated list of models the team has tested and verified. That is a separate offering, and the introductory docs page does not list prices for anything, so treat cost as “whatever your provider charges” until you have checked. One practical note: Windows users are directed to WSL for the best experience.
Verdict: worth trying if you want an open agent that is not tied to one lab or one interface. For a similar idea inside the editor, see Cline; for a terminal tool built around git commits, see Aider.
Sources: OpenCode’s documentation and GitHub repository, checked on September 21, 2026.
