Dev tools · checked 2026-10-14

Can I vibecode GitHub Copilot?

KINDA· weekend project

The visible half, an editor extension that sends the code around your cursor to a model and shows ghost text plus a chat panel, is a weekend project, and free tools such as Continue and twinny already do it with your own keys. The hard part is feel: completions must arrive quickly, know the repository and stay out of the way. Copilot's other half, review on pull requests, the cloud agent and one bill across many models, lives inside GitHub and can't be copied from outside it.

$10/moprice

$120you'd save a year

weekendbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyAdvanced · Real infrastructure or OAuth work.

TimeA weekend to get completions that feel good

Claude Code, Codex or CursorAny of them can run the prompt.
VS Code and Node.js 22The extension is built and packaged with npm.
A model to talk toOllama with a code model (a few GB of disk), or an API key for a hosted provider.
Apple Silicon or a GPUOnly if you want completions to run locally at a usable speed.

Before you buildCopilot Free already gives 2,000 completions and 50 chat requests a month. Check whether that covers you before building.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a VS Code extension to replace GitHub Copilot. Requirements:

- TypeScript for VS Code 1.95+, bundled with esbuild. `npm run package` produces a .vsix I
can install with `code --install-extension`.
- Inline completions: register an InlineCompletionItemProvider, debounce 150 ms, cancel stale
requests, and send the 60 lines before and 20 lines after the cursor as a fill-in-the-middle
prompt.
- Models: any OpenAI-compatible endpoint, set in settings (baseUrl, model, apiKeyEnvVar). Keys
are read from the named environment variable and never stored in settings. Default to local
Ollama at http://localhost:11434/v1 with a code model that supports fill-in-the-middle, for
example qwen2.5-coder.
- Context: add the open tabs' file paths and the 3 snippets from other open files that share
the most identifiers with the cursor area. Cap the prompt at 4,000 tokens, counted with
`js-tiktoken`.
- Chat: a side-panel webview that streams answers. Commands Explain selection, Fix selection
and Write tests attach the selection and its file path. Code blocks get an Apply button that
shows a diff first.
- Respect a `.aiignore` file (gitignore syntax via `ignore`): excluded files never leave the
machine.
- A status-bar item showing the model, a toggle that pauses completions, and an output channel
logging the last 50 requests with latency and token counts.
- No accounts, no telemetry. The only network calls go to the configured endpoint.
- Out of scope: pull-request review, a cloud agent, multi-file edits, JetBrains.
- Tests: vitest for the prompt builder and the debounce logic. The README lists the models you
tried and how to measure completion latency.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own GitHub Copilot.

Add commit-message generation to the VS Code extension in this folder. Requirements:

- A sparkle button in the Source Control title bar runs `git diff --staged` through the
  configured model and fills the commit input box.
- Style options in settings: imperative one-liner, or Conventional Commits.
- Truncate diffs over 6,000 tokens by file, keeping file names and hunk headers, and say so in
  the message footer.
- Respect .aiignore: ignored files are left out of the diff sent to the model.
- Update the README.

Add repository-wide context to the chat in the VS Code extension in this folder.
Requirements:

- Index every tracked text file (respecting .gitignore and .aiignore) into chunks of about
  50 lines, embedded with nomic-embed-text through Ollama and stored with sqlite-vec in
  .vscode/ai-index.db.
- Re-embed only the files whose content hash changed; show progress in the status bar.
- Chat retrieves the 8 closest chunks, includes them in the prompt and lists the file paths and
  line ranges as clickable sources under each answer.
- Out of scope: cloud vector stores.
- Update the README.

Add a command-line reviewer to the project in this folder. Requirements:

- `npx ai-review [base]` reads `git diff base...HEAD` (default origin/main) and sends it to the
  configured OpenAI-compatible endpoint in chunks of at most 4,000 tokens.
- Output Markdown grouped by file: bugs, risky changes, missing tests, each with a line number
  and a one-line reason. Say "no findings" when there are none; never invent line numbers.
- Skip lock files, generated code and anything matched by .aiignore.
- Exit code 1 when any finding is marked as a bug, so it can run in a git hook.
- Update the README.

What you lose

Pull-request review, the cloud coding agent and issue-to-PR flows on github.com
Completion latency tuned on GitHub's servers, and next-edit suggestions
One subscription spanning many models, in JetBrains, Xcode, Visual Studio and the terminal too
Seat management, policies and spending limits for organizations
Repository indexing that GitHub keeps current for you

Why people still pay

It already sits where the code lives: one login covers the editor, the pull request and the agent, and an employer can buy it centrally with policies. A flat $10 a month buys a pool of credits across several models without managing keys.

Moatintegrations execution polish

Do the math

GitHub Copilot or your own copy?

Over

Your usage

Count my time12h ×$/h

GitHub Copilot, 1 yr$120

Your own copy, 1 yr$130running + build session

Paying is cheaper by$10never pays for itself

Estimate · A local model costs nothing per request. Hosted APIs bill per token, and a heavy month can pass the price of Copilot Pro. The build is token usage on your agent plan.

Leaving GitHub Copilot

Take your data, then cancel.

1 · Take your data out of GitHub Copilot

1There is no account data to export: Copilot keeps no documents of yours.
2Your settings live in the editor, and custom instructions you wrote for a repository live in the repository itself, for example in .github/copilot-instructions.md.
3If you use a self-hosted tool next, point it at the same instruction files and copy over any prompt files you kept.

2 · Cancel GitHub Copilot

1Click your profile picture, open Settings, then Billing & licensing and Licensing (older interface: Plans and usage).
2In the GitHub Copilot section, open the Manage subscription menu and choose Cancel subscription.
3Confirm with the Cancel button for your plan (Pro, Pro+ or Max).

You keep your paid features until the end of the billing cycle, then drop to Copilot Free automatically.
You can't cancel Pro that came free as a verified student, teacher or maintainer, or through an organization: in those cases turn Copilot off in your editor instead.
If an organization assigns you a Business or Enterprise seat, your personal plan is cancelled automatically.
Remove any budget you set for extra credits so nothing is charged after you leave.

Checked 2026-10-14 against GitHub Copilot's docs: 1, 2, 3

The escape hatch

Don't feel like building it? These are already free.

ContinueOpen sourceApache-2.0 extension for VS Code and JetBrains that takes any model for chat, edits and completions, local or hosted.★ 36k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗ClineOpen sourceApache-2.0 agent for VS Code that edits files and runs commands with your approval, using your own API key.★ 70k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗Kilo CodeOpen sourceMIT coding agent extension that grew out of the Cline line of tools; bring your own provider key.★ 28k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗

All 5 free alternatives to GitHub Copilot ›

The numbers

GitHub Copilot pricing

Plan	Monthly	Annual, per month	What you get
Free	$0/user	—	2,000 inline completions and 50 chat requests a month, a small AI-credit allowance, automatic model choice only.
Pro	$10/user	—	Unlimited inline completions, 1,500 AI credits a month (1,000 base plus 500 flex), cloud agent and code review, extra credits on request.
Pro+	$39/user	—	7,000 AI credits a month (3,900 base plus 3,100 flex) and access to the full model list, including third-party coding agents in preview.
Max	$100/user	—	20,000 AI credits a month (10,000 base plus 10,000 flex), built for sustained agent work.
Business	$19	—	1,900 AI credits per user a month, with admin policies and spending limits.
Enterprise	$39	—	3,900 AI credits per user a month plus enterprise controls.

Free tier: 2,000 completions and 50 chat requests a month; no code review, cloud agent or purchased extra credits.
Billing: Monthly per user; the plan pages list no annual price. Each plan includes AI credits worth its price plus a flex bonus. When they run out you wait for the reset, set a budget for extra credits or switch to a cheaper model.
Hidden costs: Usage is metered in GitHub AI Credits (1 credit is $0.01) at each model's rate, so long agent sessions can empty a month's allowance quickly. Budget alerts fire at 75, 90 and 100 percent. Trade press reported surprise bills after the move to metered credits.
Source: docs.github.com

Also on the list

Cursor$8/mo · Dev toolsKINDA

Replit$20/mo · Dev toolsKINDA

Lovable$25/mo · Dev toolsKINDA

Questions

GitHub Copilot, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Free: 2,000 inline completions and 50 chat requests a month. Verified students, teachers and maintainers of popular open-source projects can get Pro at no cost.

Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: Pull-request review, the cloud coding agent and issue-to-PR flows on github.com, Completion latency tuned on GitHub's servers, and next-edit suggestions.

About $10 a month on the Pro plan, $120 a year (checked 2026-10-14).

Pull-request review, the cloud coding agent and issue-to-PR flows on github.com; Completion latency tuned on GitHub's servers, and next-edit suggestions; One subscription spanning many models, in JetBrains, Xcode, Visual Studio and the terminal too; Seat management, policies and spending limits for organizations; Repository indexing that GitHub keeps current for you. If any of those matter to you, keep paying.