OpenCode
Open source AI coding agent for the terminal, IDE or desktop
About OpenCode
OpenCode is an open source coding agent that runs where you already work: a terminal interface, a desktop app for macOS, Windows and Linux, or an extension inside your editor. It plans the work, edits files, runs terminal commands, checks the result and debugs itself, which is a different shape of tool from an autocomplete that suggests the next few lines and leaves the rest to you.
The part that matters most is what it is not tied to. OpenCode is model-agnostic, reaching 75+ providers through Models.dev, so you can point it at Claude, GPT, Gemini or a model running on your own machine without changing tools. Most coding agents are a front end for one vendor's model, which means your agent choice and your model choice are the same decision. Here they are separate. It also loads the right language servers automatically, so the model sees real type information and diagnostics rather than guessing from raw text. The licence is MIT, and the project offers a paid layer called Zen for people who would rather not manage provider accounts.
You install it with a single curl, npm, bun or brew command and run it in a project directory. Describe what you want changed and the agent plans the work before touching anything, then edits files, runs commands and reads the output to decide what to do next. Because it loads the appropriate LSP for the codebase, it can see type errors and diagnostics the way your editor does and correct itself instead of confidently shipping something that does not compile. You choose the model per session, so an expensive model can handle the hard reasoning while a cheap one handles mechanical edits. The same agent is available as a terminal UI, a desktop app with tabs, or through an IDE extension, and sessions can be shared via a link.
- •Model Agnostic by Design - 75+ providers through Models.dev, including local models, so the agent and the model are separate choices
- •LSP Aware - Loads the right language servers automatically so the model works from real diagnostics rather than plain text
- •Three Interfaces, One Agent - Terminal UI, desktop app for macOS, Windows and Linux, and an IDE extension
- •The Full Loop - Plans, edits, runs commands, validates and debugs itself rather than emitting single-shot suggestions
- •Zen Model Access - An optional hosted set of models the project has benchmarked specifically for coding agents
- •MIT Licensed - No usage restrictions, and the whole thing can be self-hosted or forked
Developers who live in a terminal and want an agent there rather than in a separate chat window, and anyone unwilling to bet their workflow on a single model vendor. The provider-agnostic design pays off most if you switch models often, run local models for privacy, or want to keep costs down by matching model to task. Teams with their own LLM gateway get the clearest benefit from the enterprise tier, which does not charge for tokens in that case. If you want an opinionated product where the model choice is made for you and everything is configured out of the box, a single-vendor agent will feel simpler.
Pricing
Usage based, no flat monthly plan
- Open Source$0/mo
- Zen (Pay as you go)Not listed
From the vendor pricing page, 2026-09-19
