Swap GitHub Copilot’s Built-in Model for Claude Opus 4.7: A Hands-On BYOK Guide

By default, GitHub Copilot routes through GitHub’s own model backend — meaning you don’t control which model version you’re running, and latency can be unpredictable depending on your region. Many developers want to run Claude Opus 4.7 directly for code generation and Agent tasks without paying for yet another subscription.

GitHub has officially opened up BYOK (Bring Your Own Key) support in Copilot CLI. With just four environment variables, you can point the model backend to any Anthropic-compatible endpoint — including ClaudeAPI.com. This guide covers two integration paths — Copilot CLI and VS Code’s built-in Chat view — with complete download, install, configuration, and verification steps. Follow along and you’ll be up and running in under 10 minutes.

Why the Default Copilot Setup Falls Short

Pain points you can’t avoid with the default Copilot experience:

1. Unreliable access in some regions: Both Copilot CLI and the VS Code extension route through GitHub’s own proxy — 401 errors and timeouts are common.
2. No model choice: The “Claude Sonnet” you see in Copilot? GitHub’s backend decides which version that actually is, and new model availability lags behind.
3. Agent tasks burn through your quota: Subscription quota gets consumed heavily by automated Agent calls — once you hit the ceiling, there’s no instant way to scale up.

The fix: point the model backend directly to an Anthropic-compatible relay endpoint. The API protocol is identical — you only change one URL and one key.

Endpoint used in this guide:

https://gw.claudeapi.com

https://gw.claudeapi.com

Get your API Key at claudeapi.com — sign up and start using it immediately.

1. Sign Up & Prepare

1. Visit claudeapi.com and register with your email
2. Go to the console.claudeapi.com dashboard
3. Add funds to your account
4. Create an API Key (you must select a group) and copy the key starting with sk-

Save the key in a local text editor — you’ll need it for every step below.

2. Download & Install Copilot CLI

Recent versions of GitHub CLI ship with copilot as a built-in subcommand — there’s no need to install the gh-copilot extension separately.

⚠️ Do not run gh extension install github/gh-copilot — newer versions of gh will return: copilot matches the name of a built-in command or alias

Official downloads:

• GitHub CLI homepage: https://cli.github.com/
• GitHub CLI Releases: https://github.com/cli/cli/releases
• Copilot CLI BYOK docs: https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/use-byok-models

2.1 Install GitHub CLI

Version requirement: gh ≥ 2.60 (minimum version with the built-in copilot subcommand). Verify with:

gh --version

gh --version

2.2 Verify the Copilot Subcommand Is Available

Call the built-in entry point directly — you do not need and should not run gh extension install:

gh copilot -- --help

gh copilot -- --help

You should see parameters like -p / --prompt, --stream, --no-color, --secret-env-vars in the output. If you get copilot matches the name of a built-in command or alias instead, you have a leftover legacy gh-copilot extension that needs to be removed first:

gh extension remove gh-copilot

gh extension remove gh-copilot

About subscriptions: In BYOK mode, model inference is handled by your configured custom provider (e.g., ClaudeAPI). Running gh copilot help providers explicitly states: GitHub authentication is not required when using a custom provider. In other words, your custom provider’s API Key is the prerequisite for model calls — a GitHub Copilot subscription is not strictly required. However, some advanced features (repo indexing, enterprise policies, etc.) still depend on GitHub authentication.

3. Configure Copilot CLI (Recommended Approach)

Switch to BYOK mode with four environment variables.

3.1 Field Reference

3.2 Full Environment Variable Configuration

macOS / Linux (add to ~/.zshrc or ~/.bashrc):

# === GitHub Copilot CLI BYOK: Connect to ClaudeAPI ===
export COPILOT_PROVIDER_TYPE="anthropic"
export COPILOT_PROVIDER_BASE_URL="https://gw.claudeapi.com"
export COPILOT_PROVIDER_API_KEY="sk-yourClaudeAPIkey"
export COPILOT_MODEL="claude-opus-4-7"

# === GitHub Copilot CLI BYOK: Connect to ClaudeAPI ===
export COPILOT_PROVIDER_TYPE="anthropic"
export COPILOT_PROVIDER_BASE_URL="https://gw.claudeapi.com"
export COPILOT_PROVIDER_API_KEY="sk-yourClaudeAPIkey"
export COPILOT_MODEL="claude-opus-4-7"

Apply the changes immediately:

source ~/.zshrc   # or source ~/.bashrc

source ~/.zshrc   # or source ~/.bashrc

Windows PowerShell (temporary — current session only):

$env:COPILOT_PROVIDER_TYPE = "anthropic"
$env:COPILOT_PROVIDER_BASE_URL = "https://gw.claudeapi.com"
$env:COPILOT_PROVIDER_API_KEY = "sk-yourClaudeAPIkey"
$env:COPILOT_MODEL = "claude-opus-4-7"

$env:COPILOT_PROVIDER_TYPE = "anthropic"
$env:COPILOT_PROVIDER_BASE_URL = "https://gw.claudeapi.com"
$env:COPILOT_PROVIDER_API_KEY = "sk-yourClaudeAPIkey"
$env:COPILOT_MODEL = "claude-opus-4-7"

Windows PowerShell (persistent — write to user environment variables):

[System.Environment]::SetEnvironmentVariable("COPILOT_PROVIDER_TYPE", "anthropic", "User")
[System.Environment]::SetEnvironmentVariable("COPILOT_PROVIDER_BASE_URL", "https://gw.claudeapi.com", "User")
[System.Environment]::SetEnvironmentVariable("COPILOT_PROVIDER_API_KEY", "sk-yourClaudeAPIkey", "User")
[System.Environment]::SetEnvironmentVariable("COPILOT_MODEL", "claude-opus-4-7", "User")

[System.Environment]::SetEnvironmentVariable("COPILOT_PROVIDER_TYPE", "anthropic", "User")
[System.Environment]::SetEnvironmentVariable("COPILOT_PROVIDER_BASE_URL", "https://gw.claudeapi.com", "User")
[System.Environment]::SetEnvironmentVariable("COPILOT_PROVIDER_API_KEY", "sk-yourClaudeAPIkey", "User")
[System.Environment]::SetEnvironmentVariable("COPILOT_MODEL", "claude-opus-4-7", "User")

Note: After writing to user-level environment variables, you must restart your terminal for changes to take effect.

Windows CMD (temporary):

set COPILOT_PROVIDER_TYPE=anthropic
set COPILOT_PROVIDER_BASE_URL=https://gw.claudeapi.com
set COPILOT_PROVIDER_API_KEY=sk-yourClaudeAPIkey
set COPILOT_MODEL=claude-opus-4-7

set COPILOT_PROVIDER_TYPE=anthropic
set COPILOT_PROVIDER_BASE_URL=https://gw.claudeapi.com
set COPILOT_PROVIDER_API_KEY=sk-yourClaudeAPIkey
set COPILOT_MODEL=claude-opus-4-7

3.3 Choose Your Model

Available model IDs for COPILOT_MODEL:

claude-opus-4-7
claude-opus-4-6
claude-sonnet-4-6
claude-haiku-4-5-20251001

claude-opus-4-7
claude-opus-4-6
claude-sonnet-4-6
claude-haiku-4-5-20251001

Model requirements: Must support tool calling and streaming, with a context window ≥ 128k. All four IDs above meet these criteria.

3.4 Verify Connectivity (Required)

Step 1: Hit the base_url directly with HTTP to confirm Key + endpoint are working.

macOS / Linux using cURL:

curl https://gw.claudeapi.com/v1/messages \
  -H "x-api-key: sk-yourClaudeAPIkey" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-4-7",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Reply with only pong"}]
  }'

curl https://gw.claudeapi.com/v1/messages \
  -H "x-api-key: sk-yourClaudeAPIkey" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-4-7",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Reply with only pong"}]
  }'

Windows PowerShell — use Invoke-RestMethod with a longer timeout (avoids Windows’ default 100s timeout cutting off longer responses):

$headers = @{
  "x-api-key"         = "sk-yourClaudeAPIkey"
  "anthropic-version" = "2023-06-01"
  "content-type"      = "application/json"
}
$body = @{
  model      = "claude-opus-4-7"
  max_tokens = 256
  messages   = @(@{ role = "user"; content = "Reply with only pong" })
} | ConvertTo-Json -Depth 5

Invoke-RestMethod -Method Post `
  -Uri "https://gw.claudeapi.com/v1/messages" `
  -Headers $headers `
  -Body $body `
  -TimeoutSec 120

$headers = @{
  "x-api-key"         = "sk-yourClaudeAPIkey"
  "anthropic-version" = "2023-06-01"
  "content-type"      = "application/json"
}
$body = @{
  model      = "claude-opus-4-7"
  max_tokens = 256
  messages   = @(@{ role = "user"; content = "Reply with only pong" })
} | ConvertTo-Json -Depth 5

Invoke-RestMethod -Method Post `
  -Uri "https://gw.claudeapi.com/v1/messages" `
  -Headers $headers `
  -Body $body `
  -TimeoutSec 120

Validation criteria (don’t hardcode the expected response body):

• HTTP status code 200
• The response JSON content array contains at least one element with type: "text" and a non-empty text value
• The model field matches the model ID you requested

The model’s actual text response is natural language — it could be pong, Pong!, Sure, pong., or any other variation. As long as the three criteria above are met, the endpoint + key are working correctly. Don’t set max_tokens too low (128+ recommended) — otherwise normal output will be truncated.

Step 2: Run a Copilot command:

gh copilot -- -p "Give me a single PowerShell command: find all files larger than 100MB in the current directory. No explanation."

gh copilot -- -p "Give me a single PowerShell command: find all files larger than 100MB in the current directory. No explanation."

⚠️ The legacy syntax gh copilot suggest "..." now returns Invalid command format — it has been replaced by gh copilot -- -p "...". Everything after -- is passed directly to the copilot subprocess.

Example output:

Get-ChildItem -Recurse -File | Where-Object { $_.Length -gt 100MB }

Get-ChildItem -Recurse -File | Where-Object { $_.Length -gt 100MB }

If the response is an executable command and the exit code is 0, Copilot CLI is successfully running through ClaudeAPI.com.

For a minimal connectivity self-test (no streaming, no color codes in output, API key hidden from logs):

gh copilot -- -p "Reply with only pong" --no-color --stream off --secret-env-vars=COPILOT_PROVIDER_API_KEY

gh copilot -- -p "Reply with only pong" --no-color --stream off --secret-env-vars=COPILOT_PROVIDER_API_KEY

Again, just verify: exit code 0 + stdout contains non-empty text. Don’t assert that the output must exactly equal pong.

Offline mode: In BYOK mode, Copilot CLI still communicates with GitHub’s servers by default for embeddings, repo indexing, intent recognition, etc. — only model inference goes to your custom endpoint. If you need fully offline operation, refer to the offline mode configuration at https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/use-byok-models (requires a local model).

4. VS Code Built-in Chat View (GUI Approach)

VS Code 1.94+ ships with a built-in Chat view that supports adding custom providers via Manage Models — more straightforward than the legacy Copilot Chat extension.

VS Code download: https://code.visualstudio.com/Download

Version requirement: VS Code ≥ 1.94. Check via Help → About.

4.1 Open the Configuration Panel

1. Open VS Code
2. Press Ctrl + Shift + P (macOS: Cmd + Shift + P) to open the Command Palette
3. Type and select: Chat: Manage Language Models
4. In the Language Models editor, click Add Models
5. Select Anthropic from the provider list

4.2 Fill In the Fields

After saving, go back to the Chat view (the conversation icon in the sidebar). “Claude Opus 4.7” will appear in the model selector — just switch to it.

4.3 Enterprise Accounts: Enable the Policy First

Copilot Business / Enterprise accounts require a GitHub administrator to first enable the following in GitHub.com → Organization Settings → Copilot → Policies:

Bring Your Own Language Model Key in VS Code: Enabled

Otherwise the Add Models option will be grayed out.

4.4 Don’t Confuse: The Legacy Copilot Chat Extension Doesn’t Support BYOK

Avoid this common pitfall:

If you only have the legacy Copilot Chat extension installed, uninstall it and use the VS Code built-in Chat view with BYOK instead.

5. Copilot SDK Integration (Advanced, Optional)

If you’re embedding Copilot Agent capabilities in your own tooling, configure the provider directly via the Copilot SDK:

SDK repository: https://github.com/github/copilot-sdk

import { CopilotAgent } from "@github/copilot-sdk";

const agent = new CopilotAgent({
  provider: {
    type: "anthropic",
    baseUrl: "https://gw.claudeapi.com",
    apiKey: process.env.COPILOT_PROVIDER_API_KEY!,
  },
  model: "claude-opus-4-7",
});

const session = await agent.createSession({
  systemPrompt: "You are a senior TypeScript development assistant",
});

const response = await session.send({
  message: "Convert this callback-style function to async/await: ...",
});

console.log(response.text);

import { CopilotAgent } from "@github/copilot-sdk";

const agent = new CopilotAgent({
  provider: {
    type: "anthropic",
    baseUrl: "https://gw.claudeapi.com",
    apiKey: process.env.COPILOT_PROVIDER_API_KEY!,
  },
  model: "claude-opus-4-7",
});

const session = await agent.createSession({
  systemPrompt: "You are a senior TypeScript development assistant",
});

const response = await session.send({
  message: "Convert this callback-style function to async/await: ...",
});

console.log(response.text);

The SDK also requires model support for tool calling + streaming + ≥128k context. All standard Claude 4.x models meet these requirements.

6. Troubleshooting

7. Quick Reference

GitHub Copilot CLI BYOK
├── COPILOT_PROVIDER_TYPE: anthropic
├── COPILOT_PROVIDER_BASE_URL: https://gw.claudeapi.com   (no /v1)
├── COPILOT_PROVIDER_API_KEY: sk-yourClaudeAPIkey
└── COPILOT_MODEL: claude-opus-4-7

VS Code Built-in Chat View
├── Provider: Anthropic
├── Endpoint: https://gw.claudeapi.com
├── API Key: sk-yourClaudeAPIkey
└── Model ID: claude-opus-4-7

GitHub Copilot CLI BYOK
├── COPILOT_PROVIDER_TYPE: anthropic
├── COPILOT_PROVIDER_BASE_URL: https://gw.claudeapi.com   (no /v1)
├── COPILOT_PROVIDER_API_KEY: sk-yourClaudeAPIkey
└── COPILOT_MODEL: claude-opus-4-7

VS Code Built-in Chat View
├── Provider: Anthropic
├── Endpoint: https://gw.claudeapi.com
├── API Key: sk-yourClaudeAPIkey
└── Model ID: claude-opus-4-7

Download links:

Summary

Copilot CLI’s BYOK is an official escape hatch from GitHub — four environment variables, one base_url, and Copilot runs on Claude Opus 4.7. VS Code’s built-in Chat view supports the same setup via Manage Models for a GUI-based workflow.

The end-to-end process: register for a ClaudeAPI key → install gh (newer versions include the copilot subcommand built-in — no gh-copilot extension needed) → set four environment variables → verify the endpoint with cURL / Invoke-RestMethod → confirm Copilot works with gh copilot -- -p "...".

To get started with the latest Claude models, sign up at claudeapi.com. Direct access from anywhere, no extra configuration needed.