Setup ~15 min
You need two things: a coding agent (the tool in your terminal) and a model (the brain it talks to). Both can be completely free — no paid LLM subscription required.
1 · Install OpenCode (the agent)
Pick your operating system.
-
Install OpenCode with the official script:
Copy
curl -fsSL https://opencode.ai/install | bashPrefer npm?
npm install -g opencode-aiworks too (needs Node.js 18+). -
Restart your terminal, then confirm it is on your
PATH:Copyopencode --version -
Install the file-tree plugin for terminal OpenCode:
Copy
opencode plugin --global opencode-file-treeThen restart OpenCode. The file tree appears in the sidebar.
-
Install OpenCode with Scoop (recommended) …
Copy… or with Chocolatey, or npm:
scoop install opencodeCopychoco install opencode # or npm install -g opencode-aiNo package manager? Grab the Windows binary from the releases page. Windows users often have the smoothest time inside WSL2 (Ubuntu) — if you use WSL, follow the Ubuntu tab instead.
-
Open a new terminal (PowerShell) and confirm:
Copy
opencode --version -
Install the file-tree plugin for terminal OpenCode:
Copy
opencode plugin --global opencode-file-treeThen restart OpenCode. The file tree appears in the sidebar.
-
Download the OpenCode GUI release for your OS from
github.com/anomalyco/opencode/releases.
Pick the latest stable release. Download the Windows installer (
.exe/.msi) or Linux package (.AppImage/.deb) matching your system. - Install and launch the app, then open your project folder in the GUI.
- Connect a model provider inside the app (same as step 2 below: MiMo v2.6 Flash Free, Copilot, or Ollama).
2 · Connect a free model (the brain)
Choose one. Start with the default — the fallbacks are there if signup, connectivity, or your laptop get in the way.
MiMo v2.6 Flash Free (Xiaomi) or Muse Spark 1.3 Free (Meta)
Best quality-for-effort: works out of the box in OpenCode, no API key needed.
- Open OpenCode and run
/models. - Pick your model.
-
Send a quick test prompt to confirm it responds.
CopyWhat model are you?
GitHub Copilot
Already have Copilot (free for students & OSS maintainers)? Zero setup.
- In OpenCode run
/connect, choose GitHub Copilot. - Open github.com/login/device and enter the code shown.
- Run
/modelsand pick a model.
Local model (Ollama)
Offline & no signup. Needs a capable laptop (~8 GB free RAM/VRAM).
- Install Ollama, then pull a coding model:
Copy
ollama pull qwen2.5-coder - Point OpenCode at it via
opencode.json(OpenAI-compatible endpoint):Copy{ "$schema": "https://opencode.ai/config.json", "provider": { "ollama": { "npm": "@ai-sdk/openai-compatible", "name": "Ollama (local)", "options": { "baseURL": "http://localhost:11434/v1" }, "models": { "qwen2.5-coder": { "name": "Qwen2.5 Coder" } } } } } - Run
/modelsand pick Qwen2.5 Coder.
3 · Readiness check
Navigate into any empty folder and start OpenCode:
mkdir sdd-workshop && cd sdd-workshop
opencodeType a message like "Say hello and tell me which model you are." and press Enter.
Stuck? Grab a facilitator, or ask the agent itself to help debug the setup once it responds.
4 · Optional: Browser MCP plugin
Useful for browser automation, repeatable UI checks, and screenshots during verification.
-
Prerequisite: Node.js must be installed (Node 18+ recommended).
Install instructions: nodejs.org/en/download
- Install the Browser MCP extension in Chrome or Edge:
-
Initialize the OpenCode plugin in your project:
Copy
npx opencode-browser initThis updates
opencode.jsonwith the required plugin + MCP configuration. - Restart OpenCode, then run a quick browser task in chat (for example: "Open example.com and report the page title.").