Evergreen Tools
|
← Back to AI Tools

OpenCode Open-Source Terminal Coding Agent

Run a coding agent inside the terminal, IDE and desktop you already use: OpenCode is fully open source under MIT, ships with free models yet connects 75+ providers including Claude, GPT and Gemini, and supports parallel sessions with shareable links

Tool Interface

Interactive tool will be available soon

Features

  • ✓ Genuinely open source: MIT licensed with a public repository, so you are not locked to one vendor and can self-host or use the hosted service
  • ✓ Three surfaces in one: a terminal TUI, a desktop app and an IDE extension, so you do not have to abandon your existing editor
  • ✓ Model freedom: bundled free models plus 75+ LLM providers through Models.dev, including local models; sign in with GitHub Copilot or a ChatGPT Plus/Pro account
  • ✓ Automatic LSP loading: the right language servers are wired up per project, giving the model real type and symbol information instead of guesswork
  • ✓ Parallel sessions and share links: run multiple agents on the same project at once and share a session link so a teammate can reproduce or help debug

How to Use

  1. Prepare your terminal: install a modern terminal emulator, then follow the official docs to install via the curl script, npm, brew, mise or Docker
  2. Choose your model source: start with the bundled free models, or supply your own provider API keys; existing Copilot or ChatGPT subscriptions can simply be logged in
  3. Launch from the project root so the agent can read repository context; for large changes ask for a plan before it edits any file, then approve execution
  4. Fold it into your workflow: open parallel sessions for independent tasks, share session links with teammates, and still route important changes through review and tests

FAQ

What is OpenCode?

OpenCode is an open source AI coding agent, described on its own site as the open source AI coding agent, available for the terminal, IDE and desktop. Installation options include the official install script, the opencode-ai npm package, Homebrew, mise and a Docker image. It is MIT licensed with source at https://github.com/anomalyco/opencode — see https://opencode.ai/ and https://opencode.ai/docs

How is it different from Claude Code or Cursor?

The main difference is openness and surface area. OpenCode is an MIT-licensed open source project with no single-vendor model lock-in, and it ships as a terminal interface, desktop app and IDE extension at once. Its homepage states you can connect any model from any provider, with 75+ LLM providers through Models.dev, and that local models are supported. It also does not force you to leave your editor, since it works as an IDE extension. See https://opencode.ai/ and https://github.com/anomalyco/opencode

Are the free models real?

Yes. The homepage states free models included or connect any model from any provider, so bundled free models coexist with bring-your-own-key access to Claude, GPT and Gemini. Expect rate or quota limits on free models and weaker capability than frontier models, which is fine for trial runs and light tasks. If you already pay for GitHub Copilot or ChatGPT Plus/Pro, you can log in and reuse that quota instead. Confirm current terms at https://opencode.ai/

What does automatic LSP loading actually buy me?

Language servers give an agent real type information, definitions and references. The official feature list puts it plainly as automatically loads the right LSPs for the LLM, meaning the model receives compiler-grade semantic context rather than plain-text guessing — which shows up in refactors, signature changes and cross-file references. Details at https://opencode.ai/docs

When do parallel sessions help?

When one repository has several non-conflicting workstreams, such as updating a backend interface, fixing the front-end call sites and writing tests at the same time. The official feature list includes multi-session (start multiple agents in parallel on the same project) and share links (send a link to any session for reference or debugging). Parallelism only works when task boundaries are clear and files do not overlap, otherwise runs overwrite each other. See https://opencode.ai/