What Is GitHub Copilot? Explained Simply

GitHub Copilot suggests code as you type, answers questions in a chat panel and, in its newer modes, makes multi-step changes across a project. It does this in the editor and on GitHub.com, using large language models and whatever it can see of the repository you have open.

GitHub, which Microsoft owns, launched it in 2021 together with OpenAI. These days it runs in Visual Studio Code, Visual Studio, JetBrains IDEs, Eclipse, Xcode and Neovim.

A lot of older guides still describe flat pricing and premium requests, and that no longer holds. Since June 1, 2026, every plan bills in GitHub AI Credits. Completions and next edit suggestions stay unmetered on paid plans (GitHub blog, 2026). Existing annual Pro and Pro+ subscribers are the exception and stay on premium requests until their plan expires.

Chat spends credits, and so do agent mode and the cloud agent (formerly called the coding agent). The bill now follows how much agent work a developer runs.

What Is GitHub Copilot?

Most of the product is generated text sitting next to your code. Inline suggestions, chat replies and multi-step edits all come from large language models, which get the context of your repository along with your prompt.

It’s an AI coding assistant made by GitHub, a Microsoft subsidiary. It lives in code editors, on GitHub.com and GitHub Mobile, and in the command line, and what comes out is inline completions, chat answers, multi-file edits and pull requests.

The company behind it is a code hosting platform that Microsoft acquired in 2018. The first version of Copilot was built with OpenAI.

GitHub’s own label is AI pair programmer, which puts it in with a growing set of AI pair programming tools. What sets it apart is the wiring into GitHub repositories, issues and pull requests.

Nothing compiles or tests an inline suggestion before it lands in the editor. It’s generated text, and Copilot isn’t a linter or a test runner either.

How GitHub Copilot Has Changed Since 2021

It started in June 2021 as an invitation-only autocomplete preview. By June 2026 it was a multi-model assistant with agents and usage-based billing.

The 2021 preview ran on OpenAI Codex. GitHub said Codex wrote code better than GPT-3, partly because it trained on far more public source code, and that the preview worked best with Python, JavaScript, TypeScript, Ruby and Go.

The coding agent, now called the cloud agent, followed agent mode in 2025. Both moved Copilot from suggesting lines to carrying out tasks.

On its January 28, 2026 earnings call, Microsoft reported 4.7 million paid GitHub Copilot subscribers, up 75% year over year (TechCrunch, citing Microsoft). The same outlet had reported 20 million all-time users in July 2025.

Those two numbers don’t subtract cleanly. One counts everyone who ever tried the tool, free and trial accounts included. The other counts paying subscribers six months later.

How GitHub Copilot Generates Suggestions

Copilot builds a prompt from the code around your cursor and from other open files, sends it to a large language model in the cloud and gets back a completion or a chat answer. Each keystroke or message starts the loop again with fresh context.

What Context Does GitHub Copilot Send to the Model?

What goes into the prompt is mostly whatever sits near the cursor.

• Code before and after the cursor
• The file name and programming language
• Other open tabs and neighboring files
• Comments, function names and docstrings

Inline completion works from a narrow slice of that. Chat and agent mode can pull in more files, which helps when a change spans several of them.

A full codebase never fits in the window the model reads. Opening the relevant files in tabs changes what Copilot sees, and so does closing the noisy ones.

Which Models Power GitHub Copilot?

OpenAI Codex, a model built for code, powered it at launch in 2021. Today there’s a model picker with models from OpenAI, Anthropic, Google and xAI, among others, and GitHub says those may be replaced or updated over time.

Which models you see depends on the plan and the client. GitHub Docs state that Free and Student users reach models through auto model selection only, while Pro, Pro+, Max, Business and Enterprise each have their own model lists.

▶ GIT EXPLAINED

If you commit, branch, or merge code every day, it helps to know how Git came to be. This video covers how Linus Torvalds created Git in just 10 days after a license dispute and explains how its data model actually works.

Built in 10 Days Out of Spite. Git and its data model explained in plain language.

In February 2025 the VS Code team wrote that it preferred Claude Sonnet over GPT-4o for its own agent mode work.

What Features Does GitHub Copilot Include?

The lineup covers inline code completion, next edit suggestions, Copilot Chat, agent mode, pull request summaries, code review and command line access. There’s also a cloud agent (formerly the coding agent) that works in the background and can open pull requests. What you actually get depends on the editor and the plan.

Code Completion and Next Edit Suggestions

Completion predicts the next line or block as you type and shows it as ghost text. Tab accepts it.

Next edit suggestions go by a different signal. According to the GitHub changelog from February 2025, they use your recent edits to anticipate the next change and where it belongs.

According to GitHub’s feature matrix, next edit suggestions are generally available in Visual Studio Code and Visual Studio, and still in public preview in JetBrains IDEs, Xcode and Eclipse.

Copilot Chat

Chat takes questions in plain language and answers with code, explanations or edits.

You can reach it on GitHub.com, GitHub Mobile, supported IDEs and Windows Terminal. Typical jobs are explaining code, refactoring, writing unit tests and fixing errors. Multi-file edits go through Copilot Edits in Visual Studio Code, Visual Studio and JetBrains IDEs.

Copilot CLI is a separate client for working from the terminal.

Agent Mode and the Cloud Agent

Agent mode works inside your editor. The cloud agent, which GitHub renamed from coding agent, works without you.

GitHub’s February 2025 preview description of agent mode (InfoWorld) had it iterating on its own code and fixing the errors it recognizes, with terminal commands suggested along the way.

The cloud agent runs in a GitHub Actions environment and pushes commits to a branch. It opens a pull request when you ask for one, then asks you for review. Workflows don’t run unless you approve them.

GitHub Docs list several limits worth knowing before you hand it a task.

• It makes changes only in the repository specified when the task starts
• It works on one branch at a time, with exactly one pull request per task
• A session can run for 59 minutes at most
• Repository rules the agent can’t satisfy, such as a ruleset limiting commit authors, can block it unless Copilot is added as a bypass actor
• It only works with repositories hosted on GitHub

At launch the agent was limited to Copilot Enterprise and Pro+ customers, and from June 4, 2025 each model request it made counted as one premium request (GitHub blog). It is now available on every paid plan, although Business and Enterprise admins must enable it first, and it uses GitHub Actions minutes plus AI Credits.

Model Context Protocol support lets agent mode call outside tools. GitHub’s local MCP server, for example, can search code and manage issues, and it opens pull requests too (GitHub blog, April 2025).

Pull Request and Review Tools

Copilot writes pull request summaries and reviews changes before a human does. The review feature reached general availability in April 2025 (GitHub blog).

Copilot code review is one of several AI-powered code review tools. It works best as a first pass, not a replacement for a human reviewer.

Reviews cost something too. Since June 1, 2026 each one consumes GitHub Actions minutes, which are billed on private repositories, on top of AI Credits (GitHub changelog, June 1, 2026).

Where GitHub Copilot Runs and How to Set It Up

It runs in Visual Studio Code, Visual Studio, JetBrains IDEs, Eclipse, Xcode and Neovim, plus GitHub.com, GitHub Mobile and Windows Terminal. Completions cover the editors. Chat also reaches the web, mobile and the terminal.

Which Editors and Languages Does It Support?

Feature parity is uneven. Agent mode runs in 5 of the 6 major editors, and code review in 4 of 6.

Visual Studio matches Visual Studio Code on both rows, per the GitHub Docs feature matrix, which was still in public preview as of October 2026.

Visual Studio Code is the easiest place to start. The Free plan launched there in December 2024, and the VS Code team described Copilot as a core part of the editor experience. Neovim is the odd one out, with completions but no IDE chat or agent mode.

On languages, GitHub named Python, JavaScript, TypeScript, Ruby and Go as the strongest at the 2021 preview. Others are supported too, though quality varies by language.

How to Set Up GitHub Copilot in an Editor

1. Sign in to GitHub, or create an account, and open github.com/settings/copilot to see which plan you have.
2. Install the GitHub Copilot extension or plugin from your editor’s marketplace. Recent VS Code versions include much of the Copilot experience in the editor itself, so you may only need to sign in there.
3. Sign in with your GitHub account when the editor prompts you and approve the authorization.
4. Open a code file and start typing, or write a comment describing the function you want.
5. Press Tab to accept a suggestion, or keep typing to ignore it.

Members of an organization with a Business or Enterprise plan request access on the same settings page, under Get Copilot from an organization (GitHub Docs).

Some newer models require minimum IDE or extension versions (GitHub Docs). If a model won’t show up, update the extension before you go hunting for it.

For a longer editor walkthrough, see this guide to using GitHub Copilot in VS Code.

How Much Does GitHub Copilot Cost?

Free costs $0, Pro is $10 a month and Pro+ is $39. Max runs $100, Business $19 and Enterprise $39 per user per month. Every paid plan bundles a monthly allowance of GitHub AI Credits.

Individual plans

The figures come from GitHub Docs and GitHub’s May 2026 plans announcement. Each paid allowance combines base credits that match the price with a flex allotment that GitHub says can change over time.

Max exists because users asked whether Pro and Pro+ would last under metered billing. GitHub announced it on May 12, 2026 for sustained, high-volume work. It launched on June 1 as an upgrade for existing subscribers and opened to new sign-ups after sign-ups reopened in June.

Business and Enterprise plans

Business costs $19 per user per month and comes with $19 in monthly AI Credits. Enterprise is $39 per user per month with $39 in credits, a larger pool than Business, plus additional enterprise-grade capabilities. It requires GitHub Enterprise Cloud. Knowledge bases are no longer one of the differences, because GitHub retired them on November 1, 2025.

Included credits are pooled across the organization, so lighter users offset heavier ones (GitHub blog, April 27, 2026).

GitHub added promotional credits of $30 on Business and $70 on Enterprise for June through August 2026 (GitHub blog, April 27, 2026). That window has ended, so the standing allowance now equals the seat price.

Older third-party guides say Business requires GitHub Enterprise Cloud (Swimm). GitHub Docs say organizations on GitHub Free or GitHub Team qualify too, and only Enterprise needs Enterprise Cloud. GitHub also paused self-serve Business purchases in April 2026 as a temporary reliability measure (GitHub blog, April 27, 2026), so confirm that your organization can still buy seats directly.

Neither plan is available for GitHub Enterprise Server.

How usage-based billing works day to day

Completions and next edit suggestions stay off the meter on paid plans, per GitHub’s May 2026 announcement. Everything else that calls a model spends credits.

• Chat, agent mode and the cloud agent draw from the monthly allowance
• Base credits go first and the flex allotment after, at the same rates in the IDE, on github.com and in the CLI
• Extra usage beyond the allowance needs an additional spending budget
• Existing annual Pro and Pro+ subscribers stay on premium-request billing until their plan expires, then move to Copilot Free unless they upgrade to a monthly plan (GitHub blog, April 27, 2026)

On individual plans GitHub may cap additional spend based on usage patterns, billing history and account verification (GitHub changelog, June 1, 2026). For organizations, user-level budgets limit total AI Credit usage, not only the extra spend.

Sign-ups are a moving target. GitHub paused new individual sign-ups on April 20, 2026, and its June 1 changelog said the pause remained. A June 17 changelog said sign-ups for Student, Pro, Pro+ and Max were reopening gradually over the following weeks. Plans change quickly, so confirm on the plans page before choosing.

How Does GitHub Copilot Handle Your Code and Licensing?

It depends on the plan. GitHub does not train its models on Copilot Business or Enterprise data, while interaction data from Free, Pro and Pro+ has been used for training by default since April 24, 2026 unless the user opts out.

Licensing risk is handled by a public code filter and, on Business and Enterprise, by indemnity.

Does GitHub Copilot train on your code?

Business and Enterprise data is excluded from model training, according to GitHub’s Copilot for Business FAQ. On Free, Pro and Pro+, inputs, outputs, code snippets and surrounding context feed training unless you opt out under Privacy in settings (GitHub changelog, March 25, 2026).

Guides written before spring 2026 often say individual data is not used for training by default. That changelog replaced the default.

Interaction data from members of, or outside collaborators with, a paid organization is excluded from training, and data from paid organization repositories is never used, whichever plan the user holds, according to GitHub’s FAQ as InfoQ reported it in April 2026.

What does GitHub Copilot retain?

On Business or Enterprise, prompts and suggestions from IDE chat and completions are not retained. Other access on those plans keeps them for 28 days. User engagement data is kept for two years, and cloud agent session logs for the life of the account.

Microsoft Tech Community summarizes these retention rules from GitHub’s Trust Center.

Business and Enterprise admins can also keep files out of Copilot’s context through content exclusion, generally available in IDEs since November 12, 2024 and in the Copilot app and CLI since September 2, 2026 (GitHub changelog). Coverage still varies by surface, so check GitHub’s availability table before relying on it.

How does the public code filter work?

GitHub’s earlier research put matches with public code below 1% of suggestions (GitHub blog, August 2023).

Matches turn up more often in empty or near-empty files and rarely inside an established application, per GitHub.

With the filter on, Copilot compares each suggestion and its surrounding code, about 150 characters, against public code on GitHub. A policy setting decides whether matches are blocked or shown with a reference to the source repository.

Who carries the legal risk?

Microsoft’s Copilot Copyright Commitment covers GitHub Copilot for commercial customers. Microsoft defends the customer and pays adverse judgments if a third party sues over copyright, provided the customer used the built-in content filters (Microsoft, September 2023).

The customer must also hold the rights to what it feeds into the tool.

On September 16, 2026, the Ninth Circuit affirmed the dismissal of the DMCA claim in Doe v. GitHub (No. 24-7700), holding that Copilot output is not a stripped copy of the plaintiffs’ code (Ropes & Gray and Gibson Dunn, 2026).

The panel did not decide whether stripping attribution during training could violate the DMCA, and the open-source license contract claims continue in the district court (Haynes Boone, 2026). The ruling narrows one claim without settling whether Copilot output infringes.

Does GitHub Copilot Make Developers Faster, and Where Does It Fall Short?

The evidence conflicts. GitHub’s controlled experiment found developers finished a scoped task 55.8% faster with Copilot, while METR’s 2025 trial found experienced open-source developers took 19% longer with AI tools.

The studies used different tasks and different developers, and not even the same tool.

What the research shows

• In GitHub’s experiment, the treated group finished a JavaScript HTTP server task 55.8% faster than the control group, with a 95% confidence interval of 21% to 89%. It ran from May 15 to June 20, 2022 (Microsoft Research and GitHub).
• METR put 16 experienced developers on 246 real tasks in mature open-source repositories, and they came out 19% slower (METR, July 2025).
• Veracode built test cases from 80 code completion tasks and ran them across more than 100 LLMs. In 45% of them the model introduced a vulnerability (Veracode, 2025).

METR’s main tool was Cursor, not Copilot, and Veracode tested models rather than products. Treat both as evidence about AI coding assistance in general.

Our reading of the two experiments is that they point to one variable. Greenfield work with a clear spec benefits, while work that leans on undocumented knowledge of a mature repository does not.

When GitHub Copilot is the wrong fit

Mature, undocumented codebases are the setting where METR measured a slowdown. Security-sensitive code is another weak spot, since GitHub says Copilot scans outputs for vulnerable code, but a scan is not a review.

Repositories that exceed the context window cause trouble too, because the relevant files have to be in view for suggestions to use them. Long agent runs come last. GitHub itself said agentic sessions bring much higher compute demand, which is why it moved to metered billing (GitHub blog, April 27, 2026).

Treat every suggestion like code from a new teammate. Read it, and cover it with unit tests before it goes near production.

Then send it through your regular code review process, where a human still signs off.

How Does GitHub Copilot Compare With Other AI Coding Tools, and Is It Worth Paying For?

It’s worth paying for when your code lives on GitHub and you work inside a mainstream editor. Pro covers most solo developers, and the higher tiers only pay off with sustained agent use.

Comparison with ChatGPT and other assistants

Three vendors price the entry team seat at the same level. Copilot Business and Amazon Q Developer Pro both list $19 per user per month, and Gemini Code Assist Standard lists $19 on an annual commitment and $22.80 month to month (GitHub, AWS and Google Cloud pricing pages).

At that tier, ecosystem fit matters more than price.

Cursor is the odd one out because it replaces the editor instead of plugging into it, and this comparison of Cursor against VS Code shows what that switch involves.

For a wider shortlist, see these AI code generation tools.

A decision rule for picking a plan

For completions and occasional chat, Free or Pro is enough, since completions stay unmetered on paid plans. If you run agent sessions regularly, go with Pro+ and check credit use after the first month.

Max is the tier GitHub built for all-day agent runs.

A team on GitHub should start with Business and move to Enterprise only if it needs GitHub Enterprise Cloud, its larger credit pool and the extra enterprise controls. If you’re an AWS or Google Cloud shop, trial the cloud vendor’s assistant alongside Copilot before committing seats.

Skip Max until Pro+ credits run out early two months in a row. Paying for headroom you don’t use is the most common overspend with metered plans.

GitHub Copilot FAQ

Is GitHub Copilot free for students, teachers and open source maintainers?

Yes. Verified students get the Copilot Student plan at no cost, and verified teachers and maintainers of popular open source projects may be eligible for free Copilot Pro.

Maintainers receive 12 months and can renew for free while they still maintain a qualifying project.

Does GitHub Copilot work offline?

No. Copilot sends the code around your cursor to models hosted in the cloud, so completions and chat need an internet connection.

Without one, the extension has nothing to generate suggestions from.

How do custom instructions change what GitHub Copilot writes?

A .github/copilot-instructions.md file in your repository states your conventions, and Copilot applies it to chat responses, agent work and code review. It does not apply to inline code completions.

Custom instructions became generally available in VS Code in March 2025 (GitHub changelog). Prompt files store reusable prompts in the workspace.

What happened to Copilot Workspace?

GitHub Next ended the Copilot Workspace technical preview on May 30, 2025. Pull requests created in it remain normal pull requests.

Issue-to-pull-request work now runs through the cloud agent, formerly the coding agent.

Does GitHub Copilot Enterprise still include knowledge bases?

No. GitHub retired Copilot knowledge bases on November 1, 2025 and replaced them with Copilot Spaces (GitHub changelog, August 2025).

Anyone with a Copilot account can create a Space, and Business and Enterprise admins manage Spaces through policy settings. Older guides that list knowledge bases as an Enterprise perk are out of date.

Will GitHub Copilot replace developers?

No. Even the cloud agent hands its work to a person for review, and GitHub blocks whoever requested the task from approving the resulting pull request.

Autonomous work still ends at a human sign-off.

How do you cancel or downgrade a GitHub Copilot plan?

Open your account settings, go to Billing & licensing, find the GitHub Copilot section, and choose Manage subscription, then Cancel subscription or a downgrade option.

Access continues until the billing cycle ends, after which the account drops to Copilot Free. Members who get Copilot through an organization cannot change the plan themselves, and neither can people with free access as verified students, teachers or open source maintainers.

Testing GitHub Copilot on Your Own Codebase Before You Pay

Copilot earns its cost only when it shortens work on your own repository, and no published figure can confirm that for you. A timed trial on the Free plan can.

METR measured real tasks from developers’ own repositories, while GitHub’s experiment used one predefined task. A trial built from your own tickets sits closer to METR’s design.

• Time at least five recurring tickets without Copilot
• Repeat comparable tickets with Copilot on Free
• Upgrade only if the saving survives code review

The trade-off is a billing cycle of slower adoption, and the Free plan’s monthly caps can end the test early.

This holds as of October 4, 2026. GitHub says the flex part of paid allowances can change, so rerun the credit math if the allotments move.

Teams that proceed can follow a step-by-step guide to using GitHub Copilot in daily work.

• Author
• Recent Posts

Bogdan Sandu

Bogdan Sandu is a front-end developer at TMS Outsource with 8+ years of experience in web technologies. He writes about developer tools, software platforms, and web workflows based on daily hands-on use.

Latest posts by Bogdan Sandu (see all)

• MongoDB Cheat Sheet - October 20, 2026
• Adding AI Video Generation to a Software Product: Architecture, Cost Guardrails and QA - October 20, 2026
• Show Hidden Characters, Spaces, and Line Endings in Notepad++ - October 19, 2026