OpenCode Desktop + GitHub Copilot

Why OpenCode Desktop?

OpenCode Desktop is a standalone AI coding application that integrates with GitHub Copilot. Unlike browser-based tools, it runs on your machine and can connect to multiple AI providers — including GitHub Copilot, which gives you access to powerful models like Claude 3.5 Sonnet and GPT-4o for coding tasks.

Best for: Students who want a dedicated AI coding interface with access to top-tier models, without needing VS Code or other IDEs.

---

What you need before starting

Requirement	Why
GitHub account	Required for Copilot access — create one free at github.com
GitHub Copilot access	Free for verified students via GitHub Education
OpenCode Desktop	The AI coding application — download below
10 minutes	Full setup from scratch

GitHub Education gives you Copilot Pro

As an XJTLU student, you can get GitHub Copilot Pro for free while enrolled. This unlocks the coding agent, premium models, and unlimited usage. Apply at education.github.com — approval usually takes a few hours.

---

Step 1 — Install Git or GitHub CLI

OpenCode Desktop needs Git to access your repositories. You have two options:

Option A: GitHub CLI (Recommended for beginners)

GitHub CLI (gh) includes Git functionality plus GitHub-specific commands. It is simpler to use and has fewer edge cases.

Windows:

winget install GitHub.cli

Or download from: cli.github.com

macOS:

brew install gh

Linux:

sudo apt install gh  # Debian/Ubuntu
sudo dnf install gh  # Fedora

Verify installation:

gh --version

Authenticate with GitHub:

gh auth login

Follow the prompts:

1. Choose GitHub.com
2. Choose HTTPS as protocol
3. Choose Login with a web browser
4. Copy the one-time code, press Enter to open browser
5. Authorize GitHub CLI
6. Return to terminal — it should confirm success

---

Option B: Git (Full version control)

Install Git if you want the complete version control system — this is what professionals use.

Windows:

• Download from: git-scm.com
• Run installer, accept defaults
• Git Bash will be available in your right-click menu

macOS:

xcode-select --install

Or: brew install git

Linux:

sudo apt install git  # Debian/Ubuntu
sudo dnf install git  # Fedora

Verify installation:

git --version

Configure your identity (required):

git config --global user.name "Your Name"
git config --global user.email "your.email@xjtlu.edu.cn"

Authenticate with GitHub:

gh auth login  # GitHub CLI works for Git auth too

---

Step 2 — Install OpenCode Desktop

Download: Go to opencode.ai and download the desktop application for your OS.

Install:

• Windows: Run the .exe installer
• macOS: Drag to Applications folder
• Linux: Use the .AppImage or package manager

First launch: Open OpenCode Desktop. You will see a clean interface with a chat panel and project browser.

[Screenshot placeholder — OpenCode Desktop initial screen]

---

Step 3 — Connect GitHub Copilot

This is where OpenCode Desktop becomes powerful — you get access to premium AI models through Copilot.

In OpenCode Desktop:

1. Click the Settings icon (⚙️) in the bottom-left
2. Go to AI Providers or Models tab
3. Find GitHub Copilot in the list
4. Click Connect or Authenticate
5. A browser window opens — sign in to GitHub if prompted
6. Authorize OpenCode to access Copilot
7. Return to OpenCode — you should see “Connected”

[Screenshot placeholder — Settings → AI Providers with GitHub Copilot connected]

Select your model:

• Claude 3.5 Sonnet — excellent for complex reasoning, architecture, and explanations
• GPT-4o — strong all-rounder, good for code generation
• Copilot default — uses GitHub’s optimized model

Model choice matters

Different models excel at different tasks. For this course:

• Architecture planning → Claude 3.5 Sonnet
• Code generation → GPT-4o or Copilot default
• Debugging → Claude 3.5 Sonnet
• Refactoring → GPT-4o

Try the same task in two models and compare — record findings in your Dev Log.

---

Step 4 — Open or clone your project

If your repository exists on GitHub:

1. Click Clone Repository in OpenCode Desktop
2. Paste your repository URL (e.g., https://github.com/yourteam/yourproject)
3. Choose a local folder
4. Click Clone

OpenCode will use your GitHub authentication automatically.

If starting fresh:

1. Click New Project
2. Choose a folder on your machine
3. OpenCode will initialize a Git repository

---

Step 5 — Start coding with AI

OpenCode Desktop works through conversation. You describe what you want, it writes the code.

Good prompt patterns:

I am building a cat diary PWA for XJTLU volunteers.

User story: As a volunteer feeder, I want to record
that I fed Xiaobai at 7pm with 100g of dry food,
so that other volunteers can see she has been fed.

Stack: React + Vite + Supabase

Create a form component with:
- Cat selector dropdown (fetch from /api/cats)
- Time picker (default to now)
- Food type and amount fields
- Submit button

Include input validation and error handling.

What OpenCode will do:

1. Analyze your request
2. Show a plan (optional — you can ask it to think first)
3. Write the code across multiple files if needed
4. Explain what it wrote
5. Wait for your approval before saving

[Screenshot placeholder — OpenCode Desktop showing code generation]

---

Using the Coding Agent

If you have GitHub Copilot Pro (free for students), OpenCode Desktop can access the Coding Agent — an autonomous AI that works on entire features.

How to use:

1. Open your project in OpenCode Desktop
2. In the chat, type: @agent or click the agent icon
3. Describe the full task with acceptance criteria
4. The agent will:
   • Create a plan
   • Edit multiple files
   • Run terminal commands
   • Test the output
   • Report back when done

Example agent prompt:

@agent Build a feeding log feature for our cat diary app.

Acceptance criteria:
- Volunteer can select a cat from dropdown
- Form records: cat name, time, food type, amount (grams)
- After submit, confirmation message appears
- Entry appears in the cat's feeding log immediately

Stack: React + Supabase
Database table 'feeding_logs' already exists.

Do not change: the cat profile page or existing styles.

Review before committing

The coding agent is powerful but not perfect. Always:

1. Read every file it changed
2. Understand what each function does
3. Test the feature yourself
4. Only commit when you can explain it

Don’t take on features that are beyond your understanding just because you think the agent can handle it. If you cannot follow what it produced, the task was too big — shrink the scope, not your responsibility.

---

Workflow comparison: OpenCode vs other tools

Task	OpenCode Desktop	GitHub Copilot in VS Code	Browser-based (Claude, ChatGPT)
Inline completions	✗ Chat-based	✓ As you type	✗ Paste/copy
Multi-file editing	✓ Full agent mode	✓ Agent mode	✗ Manual
Model choice	✓ Multiple providers	✓ Copilot models only	✓ Any model
Standalone app	✓ No IDE needed	✗ Requires VS Code	✓ Browser only
Works in China	✓ With Copilot	✓ Via VS Code	Varies

---

Troubleshooting

Problem	Solution
Copilot not showing in providers	Make sure you have Copilot Pro (free for students) — check at github.com/settings/copilot
”Not authenticated” error	Run gh auth login in terminal, then restart OpenCode
Model responses are slow	Try switching to Copilot default model — Claude 3.5 Sonnet is powerful but can be slower during peak times
Code does not work	Rewrite your prompt with more specific acceptance criteria — vague tasks produce vague code
Cannot clone repository	Check that you are logged in to GitHub in OpenCode Settings → Accounts

---

When to use OpenCode Desktop

Good fits:

• You want a clean, focused AI coding interface without IDE complexity
• You need access to multiple AI models (Claude, GPT-4) through one tool
• You are working on well-defined features with clear acceptance criteria
• You want autonomous agent mode for scaffolding new features

Consider alternatives if:

• You already love VS Code and want inline completions → use GitHub Copilot extension
• You need heavy refactoring across many files → try Cursor (excellent multi-file editing)
• You are prototyping UI from images → try TRAE (image-to-code specialist)

See AI Coding Tools for the full comparison.

---

Your Dev Log is your evidence

When you use OpenCode Desktop (or any AI tool), record:

• What task you gave it
• Which model you used
• What worked, what needed fixing
• What you learned about prompting

This is evidence of your AI collaboration skills — assessors look for this in Week 5 onwards.