Free AI Agent for VS Code: Which One to Install in 2026
MarkdownCopilot, Cline, Antigravity, and local models through Ollama: a free start, limits, and working with sensitive code.
Read full article
A couple of years ago, AI in an editor could do one thing: suggest the next line while you wrote the code yourself. Now it takes on the whole task, rummages through files on its own, edits them, and runs checks, while you just watch and say “yes” or “redo it.” That kind of helper inside an editor is called an AI agent, and the most common question about it is simple: can you install one in VS Code for free, and which one should you choose?
Here’s the answer, in order and without fluff. The main point up front, so we don’t drag this out: there are free options, several of them, and they differ in what they offer. Below: how they differ, which is easiest for a beginner to install, and where your code doesn’t go to someone else’s cloud. Everything is based on official documentation from GitHub, Google, Cline, and Ollama; all the links are at the end.
Practical instructions for using Ailoot are in the documentation.
Can I use an AI agent free of charge in VS Code, and what is included?
Key takeaway
It’s an AI assistant that lives right inside the VS Code editor and doesn’t just suggest lines of code: it completes tasks itself. It reads your files, edits them, and runs commands. “Free” here means you can start using it without a card or payment, though every free option has a usage ceiling. You install it either as an extension inside VS Code or as a separate program alongside it.
VS Code is Microsoft’s free code editor, and most vibe-coding tools work in and around it. It doesn’t include AI on its own; you add AI on top. It’s like an empty kitchen: the stove and counter are there, but you hire the cook separately.
An agent differs from a simple suggestion by seeing the task through to the end on its own. You say, “add a request form to the page,” and it finds the right files, makes the changes, and checks that nothing broke. If you’d first like to understand the idea of an agent without tying it to specific tools, there’s a separate explainer on what an AI agent is, in simple terms.
The word “free” comes with a caveat. Free almost always means “free up to a certain limit”: a set number of actions per month, a weaker model, or restrictions you have to live with. There’s only one way to get completely unlimited use, and we’ll get to it. For now, keep in mind: every option below offers a free start, but none is unlimited.
Free extension and free model: what’s the difference?
Key takeaway
These are two different things, and this is where people most often get tripped up. The extension is the agent itself, the box you install in VS Code. The model is the brain the agent consults for every answer, and using that brain costs money. The box can be free and open source while the brain costs money. It’s truly free only when both parts are free at the same time.
This is where beginners most often get confused, so let’s go into more detail. Many popular agents, such as Cline and Continue, are free and open source in themselves. You install them in VS Code and don’t pay for the extension. But they’re empty inside: they don’t have their own neural network, so they go to someone else’s model for every answer. A model’s work is measured in tokens, pieces of text you pay for, and you’re the one who pays for those.
Here’s the formula worth remembering:
Free agent = free extension + free model source.
If only the box is free, but you connect its brain with a paid key, then in practice you’re paying. You’re billed for the model’s tokens. There are three real ways to get both for free at the same time: GitHub Copilot’s built-in limit, Google Antigravity’s free plan, and your own model on your own computer through Ollama. We’ll cover each one below.
Understanding what gets passed to an agent to work on is exactly what context engineering is about.
GitHub Copilot Free: the easiest start without a key?
Key takeaway
Yes, it’s the easiest way in for a beginner. Copilot is GitHub’s official assistant. It installs in VS Code in one click and only asks you to sign in with a GitHub account; no keys or setup required. The free plan gives you two thousand code completions per month, plus chat. The model is selected automatically, and you can’t choose a top-tier one manually, but that’s more than enough to try an agent in action.
If you’ve never installed AI in an editor before and want to get started today without any hassle, start here. GitHub and Microsoft make GitHub Copilot, just as they do VS Code itself, so it’s the easiest to set up: open the extensions marketplace, install Copilot, sign in with your GitHub account, and you’re done. No need to get a key or think about tokens.
According to GitHub’s documentation, here’s what the free plan offers:
The Copilot Free plan includes up to two thousand code completions per month.
The free plan also includes chat, where you can ask questions, get an explanation of a piece of code, or ask for a rewrite. It also has agent mode, where Copilot edits files and runs commands on its own, but the free plan marks it as limited. The model on this plan is selected automatically; you can’t manually choose the top-tier one. That’s the trade-off for using it free.
Who it’s for: anyone who wants to try the idea without wrestling with settings. It has one clear downside: data goes to GitHub and Microsoft’s servers; nothing runs locally. That’s no big deal for a practice task, but it’s something to consider with sensitive code. We’ll discuss that in the section about Ollama.
Cline: a free agent, but do you pay for the model?
Key takeaway
Cline is a popular open-source agent, and the extension itself is free. But it doesn’t have a model built in, so you have three options: connect your own paid key and pay for tokens, use free models from its built-in collection, or connect your own local model. Its strength is a mode where Cline shows you a plan first and changes files only after you say “yes.”
Cline is an open-source agent for VS Code, exactly the “free box, separate brain” case we discussed above. Its developers describe it this way:
Your AI agent for complex work: reads files, writes code, runs commands—and does it all with your permission.
The extension itself is open source and free. The question is where to get the model, and Cline gives you three options. First, connect your own key from a cloud provider and pay them directly for tokens. Second, use built-in access, which includes free models. Third, the most interesting option if you’re trying to save money: connect a local model on your computer. The developers say this directly:
Ollama and LM Studio—run local models right on your machine.
Another advantage of Cline is its mode where it shows you a plan first and touches files only after you confirm. For a beginner, that’s a safeguard: you can see what the agent is about to do and stop it if it’s about to do something silly, before it breaks anything. Cline becomes completely free when paired with a local model, which we’ll cover in the next section.
Google Antigravity: what replaced Gemini Code Assist?
Key takeaway
Google used to offer a free assistant called Gemini Code Assist right inside VS Code. In 2026, Google discontinued it for individuals and replaced it with Google Antigravity on June 18. Antigravity is free on the personal plan, gives access to powerful models, and offers unlimited code completions. The catch is that it’s a separate program alongside your editor; it doesn’t install inside your VS Code. It’s important to understand that from the outset.
If you’ve read older recommendations and searched for “Gemini Code Assist for VS Code,” you should know right away: that option is gone. Google says so itself:
As of June 18, 2026, the Gemini Code Assist IDE extensions stopped serving requests for the individual plan, as well as Google AI Pro and Ultra. The same applies to Gemini CLI.
Google Antigravity took its place, and its free plan is generous. According to the pricing page, the personal plan costs zero per month and gives you access to powerful models (including third-generation Gemini and Claude Opus 4.6), unlimited code completions, and unlimited commands. There is a limit, but it’s lenient: a weekly baseline limit for heavy actions, with no exact figure on the website.
There’s one caveat that would make this recommendation misleading if left out. Antigravity doesn’t install as a plugin inside your VS Code. It’s a separate program, its own editor built on the same foundation as VS Code, but with an agent at its center. Google calls it an agentic development environment. In practice, you install an entirely different editor. If staying in your own VS Code with a plugin is essential, this isn’t the option for you. If you don’t mind switching to a separate app for a powerful free agent, it’s a good choice. For more about the tool itself, see the review of Google Antigravity.
Cline and Ollama: how can you get a free, offline agent?
Key takeaway
This is the only truly unlimited, free option. Install Ollama, an engine for running neural networks right on your computer, download a model, and connect it to Cline or Continue in VS Code. With a downloaded local model, inference can work without the internet and has no provider token quota. The extension, telemetry, and network-capable tools have separate data flows; local inference alone does not make the entire agent offline or private. The only cost is that you need a decent computer: weak hardware can run only small models.
Here’s the option where both parts are free at once, with no ceiling. Ollama is an open-source engine that runs neural networks locally on your own hardware. From its website:
Work with open models on your computer. Get started for free. Fully offline for important tasks.
Ollama itself is an engine for running models; it doesn’t have its own VS Code extension, so you connect it to the editor through an agent. Here’s how to put it together:
- Install Ollama. Download the engine from the website and install it like a regular program.
- Download a model through it. For example, Qwen or gpt-oss; both are open and free.
- Install an agent in VS Code. Cline or Continue will work; both support local models.
- Connect the model to the agent. In the agent’s settings, select Ollama as the source and specify the model you downloaded.
You end up with an agent that lives in your editor and runs on a model from your own computer.
Ollama supports a wide range of open models:
Run Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma, and other models.
The main advantage is local inference: prompts can stay on your computer when Cline points to your own local Ollama endpoint and a downloaded local model. Confirm those settings for both planning and acting, disable cloud providers or fallback and optional telemetry, and separately restrict browser, MCP, and shell tools that can access the network. Test a harmless task with external network access blocked before using sensitive code; local inference does not by itself prove complete privacy. The honest trade-off is that you need a decent computer. A good graphics card or lots of RAM can run a larger, smarter model. Weak hardware limits you to small models that are noticeably less capable. So this is the choice for people with a computer that can handle it and who need their data to stay at home.
What about Continue and Roo Code?
Key takeaway
People often ask about them because they show up in older articles. Honestly, neither is in good shape anymore. Roo Code shut down on May 15, 2026, and its repository is frozen; the developers recommend going back to Cline. Continue has also moved to “read-only” mode: you can still install and use it, but it won’t receive updates. There’s no reason to install either from scratch in 2026.
If you’ve googled Roo Code or Continue, check their current status first, or you might waste time on a dead tool.
Roo Code was a popular fork of Cline, but it shut down in 2026. The developers wrote:
The Roo Code extension was shut down on May 15. If you’re looking for a replacement, check out Cline, from which Roo Code originally grew.
Continue is still around, but development has stopped. Its project page says the repository is no longer actively maintained and is available only in read-only mode, and the last extension release was the final one. You can still install it, and it can even work with a local model through Ollama, but there’s little point in choosing a frozen tool when Cline is still actively developed.
The conclusion for both is simple: in 2026, beginners should choose actively maintained options. Skip Roo Code; use Continue only if it’s already set up and working for you. If you want a fresh overview of the tools available, see the guide to what to code with in 2026 and the roundup of free models for coding.
Which AI agent should you choose for your situation?
Key takeaway
A quick guide to the options. Want to get started today without any hassle? GitHub Copilot Free installs in one click and doesn’t need a key. Need a powerful free agent and don’t mind switching to a separate program? Google Antigravity. Privacy matters and you have a decent computer? Cline with a local model through Ollama. All three are genuinely free; they just suit different needs.
There’s no single right answer here. It depends on what matters most to you. Here’s everything in one table for a quick comparison:
| What matters most to you | What to choose | Why |
|---|---|---|
| Get started today, with no setup | GitHub Copilot Free | One click in the marketplace, sign in with GitHub, no key needed |
| A powerful free agent | Google Antigravity | Zero per month, powerful models, but a separate program |
| Local inference without a provider token quota | Cline + Ollama | A downloaded local model runs on your computer; check telemetry, fallback, and tool network access separately |
| An open-source agent with your own key | Cline | The extension is free; you pay only for model tokens |
Here’s the recommended order. If you’re just trying this out and have never installed AI in an editor, choose Copilot Free and get used to the feeling of having an agent do the work for you. When you want something more powerful for free, try Antigravity, but know in advance that it’s a separate app. And if you reach the point where privacy and unlimited use matter, install Ollama and connect a local model to Cline. A professional stack built around VS Code and Claude Code is a paid option; that’s a topic for another conversation. If you want to compare paid and free options side by side, see the guide to Claude Code alternatives for your use case.
Frequently asked questions about free AI agents for VS Code
Key takeaway
Quick answers to the most common questions. Free agents really do exist, but every free option has a limit. You don’t need to know how to program to install them. A local model can keep inference on your machine; privacy also depends on telemetry, fallback, and tool access. And even a free agent can quietly spend your money if you connect a paid key, so keep an eye on that.
Frequently asked questions
Can you install an AI agent in VS Code completely for free?
Yes. There are three genuinely free options: GitHub Copilot Free (two thousand completions per month plus chat, no key needed), Google Antigravity on a personal plan that costs zero per month, and a local model through Ollama paired with Cline. The first two have a monthly or weekly ceiling. The Ollama option has no limits at all, but you need a decent computer.
What’s the difference between a free extension and a free model?
The extension is the agent itself, the box in VS Code. The model is the brain the agent consults for every answer, and using that brain costs money. Cline and Continue are free and open source themselves, but don’t include models: connect your own paid key and you pay for tokens. It’s truly free only when both parts are free: Copilot Free, Antigravity, or a local model.
Do you need to know how to program to install an agent?
No. Copilot and Antigravity require only signing in to an account, with no setup at all. Cline and Continue can also be installed from the extensions marketplace in a couple of clicks; connecting a model is the slightly trickier part. Installing an agent takes a couple of clicks and signing in to an account. You don’t need to know how to program.
Which option is safer for sensitive code?
A downloaded local model through Ollama can keep inference on your computer. Verify the local endpoint and model, disable cloud fallback and optional telemetry, and restrict network-capable tools separately. Copilot, Antigravity, and Cline with cloud keys use remote model services. A local model alone does not prove that no data leaves the agent system.
Is it true that Gemini Code Assist no longer works in VS Code?
For individuals, yes: the old free extension is gone. Google discontinued Gemini Code Assist for personal use and replaced it with Google Antigravity on June 18, 2026. Antigravity is free on the personal plan, but it runs as a separate program alongside the editor; it doesn’t install as a plugin in your VS Code. If you see an old guide about Gemini Code Assist, it’s out of date.
Sources
- GitHub documentation, “GitHub Copilot plans”
- GitHub documentation, individual plans and billing
- GitHub Copilot in the VS Code documentation
- Cline documentation, “What is Cline”
- Cline project README on GitHub
- Google for Developers, Gemini Code Assist overview
- Google for Developers, on the discontinuation of Gemini Code Assist for individuals
- Google Antigravity pricing page
- Roo Code project page on GitHub
- Continue project page on GitHub
- Ollama website
- Ollama project README on GitHub
You can continue getting to know Ailoot at ailoot.ai.
More details
Questions about Free AI Agent for VS Code: Which One to Install in 2026
Does connecting Ollama make the whole agent private?
It keeps inference local only when the selected model and endpoint are local and cloud fallback is disabled. Telemetry, MCP servers, browser tools, and shell commands can still send data elsewhere; review or disable them separately.
How can I check the setup before using sensitive code?
Verify that the agent points to your local Ollama endpoint and a downloaded local model. Disable cloud providers or fallback and optional telemetry, restrict network-capable tools, then test a harmless task with external network access blocked.