GitHub Copilot: what it is and how it differs from AI agents

GitHub Copilot is an AI assistant for developers from GitHub that works in the editor and on GitHub itself: it suggests code as you type, answers questions about the project and can carry out multi-step tasks. If you have only heard the name and want to understand what it does and how it differs from Cline, Cursor and Claude Code, this overview is for you. We also show where your own API key fits in the picture.

The description below relies on GitHub and VS Code documentation as of October 18, 2026. Copilot plans, limits and prices change, so we do not quote them: see the current plan on GitHub's site.

What GitHub Copilot is

By the definition in GitHub's documentation, Copilot is an AI assistant that helps you write, understand and ship software. It suggests code as you type, answers questions about a codebase, reviews your changes and works on tasks you assign it. Because Copilot is built into GitHub, it works alongside what you already have: code and its history, issues, pull requests, automations in the repository.

It is not a single model but a set of features on top of which different language models run. Which models are available depends on your plan and your organization's policies.

What Copilot consists of

GitHub's documentation splits its capabilities into four groups.

Assistive. Copilot reacts to what you do: it suggests code as you type (inline suggestions), answers in chat, explains fragments. You review and apply every suggestion yourself.

Agentic. You describe a goal and Copilot goes through several steps: it explores the repository, proposes a plan, edits files, reviews pull requests, runs tools and prepares the result for your review. Responsibility for checking stays with you.

Customizations. Instructions, reusable prompts, repository context, external tools, skills and custom agents. The better Copilot understands your project's structure and conventions, the more accurate the answers.

External agents, models and tools. Agents, models and MCP servers from other providers can work alongside Copilot.

Practical takeaway: the word "Copilot" covers autocomplete in the editor, chat and an agent that carries out a task in several steps. When someone says "Copilot is not working", it is worth clarifying which of the features they mean.

Where it can be used

Copilot is available in editors (VS Code and other IDEs), on the GitHub website and on the command line (Copilot CLI): the full list depends on your plan and client, see the "Where to use GitHub Copilot" page in GitHub's documentation. In VS Code it works as a built-in feature: chat and agents launch from the sidebar, suggestions appear right in the text. Detailed setup is described in the VS Code and GitHub documentation.

How Copilot differs from Cline, Cursor and Claude Code

The differences are easiest to see through three questions: where the tool lives, whose model it uses and who sends the bill, and whether there are suggestions as you type.

What to understand here.

Copilot is the only one on the list with a suggestion-as-you-type service tied to a GitHub account. The others are mostly about chat and agents. If you are used to code completing itself, that is a strong argument for Copilot.

Cline and Kilo Code are built around your key from the start: you pick a provider, enter a Base URL and a key, and pay for tokens directly. Connecting each is described on the VS Code extensions↗ page. Roo Code, often mentioned alongside them, was shut down on May 15, 2026 according to its developers, so we do not consider it for new projects.

Cursor is a separate editor with its own interface: to connect your own key, see our guide to Cursor↗ and Cursor's own documentation. We do not quote subscription terms and prices.

Claude Code is a terminal agent that works with a project from the command line and also has a VS Code extension. Its connection is described in the guide↗. A detailed comparison is in GitHub Copilot vs Claude Code↗.

We will not claim that any of these tools "writes code better". It depends on the model, the task and your project: try them on your own repository.

Where your own key fits

You connect your own key in three scenarios.

1. In VS Code through the built-in Bring Your Own Key (BYOK) mechanism. According to the VS Code documentation, such models work in chat without GitHub sign-in and without a Copilot plan, but suggestions as you type, semantic search and embedding-based features remain with the Copilot service.
2. In the Cline and Kilo Code extensions. The key is the main way of working.
3. In terminal agents like Claude Code, OpenCode and Codex.

A Zukko fetch_... key (created in the dashboard↗) fits all three. In VS Code, for example, it looks like this (Custom Endpoint, Chat Completions type):

[
  {
    "name": "Zukko",
    "vendor": "customendpoint",
    "apiKey": "${input:zukkoApiKey}",
    "models": [
      {
        "id": "claude-sonnet-5-5",
        "name": "Claude Sonnet 5.5",
        "url": "https://api.zukko.pro/v1/chat/completions",
        "toolCalling": true,
        "vision": true,
        "contextWindow": 1000000,
        "maxOutputTokens": 32000
      }
    ]
  }
]

You pay per token, with no subscription. For reference, as of October 18, 2026 one million input and output tokens cost $0.4 and $2 on claude-sonnet-5-5, $0.27 and $1.32 on gpt-6.1-sol, and $0.013 and $0.038 on deepseek-v4.1-flash in Zukko. Example: a session of 30 requests of 6000 input and 900 output tokens on claude-sonnet-5-5 without cache is 30 × 6000 = 0.18M input and 30 × 900 = 0.027M output, that is 0.18 × 0.4 + 0.027 × 2 = 0.072 + 0.054 = $0.126. In real agent work the context grows, so set limits on the key in the dashboard in advance. The current list of models and prices is GET https://api.zukko.pro/v1/models.

Where to start

• If you need suggestions as you type and you can arrange access to Copilot, start with it: enable the features in VS Code following the guide.
• If an agent that reads the project and edits files by itself matters more and you want to pay per token, try Cline or Claude Code on your own key.
• If you are in Russia, access to models goes through a gateway: payment by SBP, MasterCard or crypto.

FAQ

What is GitHub Copilot in plain words?

An AI assistant for programmers from GitHub: it completes code as you type, answers questions in chat and, in agent mode, carries out tasks in several steps.

Is Copilot a model?

No, it is a set of features and a service around language models. Which models are available to you is determined by your plan and your organization's policies.

How does Copilot differ from Claude Code?

Copilot is built into the editor and GitHub and gives suggestions as you type, while Claude Code is a terminal agent that works with a project from the command line and is paid through your key or an Anthropic account. They solve adjacent but different tasks.

Can I use Copilot with my own key?

In VS Code, chat with your own model (BYOK) works without a Copilot plan. Suggestions as you type remain with the Copilot service.