Can I use AI in the free version of Zed?

Yes. The Personal plan is permanently free and includes 2,000 accepted edit predictions. It also allows you to bring your own API keys from providers like Anthropic or OpenAI to use the Agent features, meaning you have complete control over your AI expenses.

Can Zed replace VS Code?

It depends on how heavily you rely on extensions. For pure coding, running LSPs, and using Git, Zed is fully capable and much faster. However, if you heavily depend on specific VS Code extensions—such as debuggers for niche languages—you may hit a roadblock for now. We recommend testing it out on a side project for two weeks.

What should I do if the Agent Panel messes up my code?

All changes go into a review area. Press Ctrl+Shift+R to open the multi-buffer review page, where you can accept or reject changes section by section. Getting into the habit of checking the diff before hitting accept will save you much more time than doing a git reset later.

What are the keyboard shortcuts for edit predictions?

By default, Alt+Tab is used across all platforms to accept a prediction, and Escape cancels it. Alt+K accepts up to the next word boundary, and Alt+J accepts to the end of the line. You can customize these shortcuts in the keymap editor.

Zed Editor Guide: From Installation to Using the Agent Panel for Full-Project AI Edits

Written in Rust, Zed is a blazing-fast editor with fully built-in AI features. This guide walks you through installation, setting up API keys, and editing prediction shortcuts, before diving into how to add context in the Agent Panel, review every line of AI-generated code, and wrap up with crucial cost and security considerations for Taiwanese engineers.

A firmware engineer friend of mine in Hsinchu complained to me last week: he opened his company's massive 100,000+ line repository, and the editor spent a full forty seconds spinning its wheels. "I open it three times a day, which means I spend nearly ten hours a year just watching a loading circle."

That sentence is pretty much the entire reason Zed exists.

What is Zed?

Zed is a code editor written from scratch in Rust by the original team behind Atom, built with one main priority: speed. Latency for opening files, global searches, and scrolling through long files has been pressed down to an absolute minimum, making the difference especially noticeable on large projects. It is also open-source and features built-in real-time multi-user collaboration—not as a plugin, but right in the core.

The AI capabilities weren't just slapped on afterward either: edit predictions, the inline assistant, and the Agent Panel form a three-tier system deeply embedded in the editor. Alternatively, you can turn them off completely and just use it as a purely fast code editor.

What Can It Do for You?

Type-ahead completion: Edit predictions guess what you're about to change next based on your recent edits, going well beyond just completing the current line.
Select and rewrite: The inline assistant lets you highlight a block of code and directly issue commands to rewrite it.
Hand off full tasks: The Agent Panel can read and write project files, run tools, and modify multiple files, leaving you to review the changes section by section when it's done.
Generate commit messages: The Git panel can automatically generate commit messages for you.
Bring your own model: Zed supports hosted models, your own API keys, and local models (via Ollama).

How to Use It: Six Steps

Step 1: Installation

Head over to zed.dev to download it; versions are available for macOS, Linux, and Windows. macOS and Linux have higher maturity, while Windows support started more recently. When you open it for the first time after installation, it will ask if you want to import your VS Code settings and keybindings—I recommend saying yes, as it makes the transition much smoother.

Step 2: Decide on AI Pricing

This is the first thing you should figure out, and you have three paths:

Completely Free: Use the Personal plan, which includes 2,000 accepted edit predictions.
Bring Your Own API Key (the one I chose): The free tier lets you plug in your Anthropic, OpenAI, or Google keys, meaning you only pay for exactly what you use.
Pro Subscription: $10 per month, which includes $5 in model credits and unlimited edit predictions, with usage beyond that billed at API rates plus a 10% markup.

The beauty of bringing your own key is especially apparent for freelance developers: in months without projects, your expense is literally zero.

Step 3: Configure Edit Predictions

Press Cmd+, (or Ctrl+, on Windows/Linux) to open the settings editor, search for "Edit Predictions," and click Configure Providers. Aside from the default Zed AI, you can also select GitHub Copilot, Codestral, or hook up Ollama to run local models.

For actual usage, memorizing just these shortcuts is plenty:

Alt+Tab: Accept prediction (cross-platform)
Escape: Cancel
Alt+K: Accept up to the next word only
Alt+J: Accept up to the end of the line

That "accept a partial prediction" feature is extremely practical—it's very common for AI to guess the first half right and the second half wrong.

Step 4: Open the Agent Panel

Use the command palette to type agent: new thread, or simply click the ✨ icon in the bottom-right corner of the status bar. Once the panel opens, it presents a chat interface where you can type your request and hit Enter.

Step 5: Feed It Context

This step dictates whether you succeed or fail. Use the @ symbol to mention files, folders, symbols, or even previous chat threads to feed them into the context:

@src/lib/auth.ts There is a bug in the session expiration logic in this file,
causing users across different time zones to log out prematurely. Please find and fix it.

You can also paste code directly (Zed will automatically format it into context), or drag and drop images into the chat for vision-capable models to review.

If you want to follow along and watch the AI modify code in real-time, click the crosshair icon in the bottom-left corner of the panel, and the editor will automatically jump to the file it's currently working on.

Step 6: Review Every Line of Changes

Once the AI finishes making changes, the panel will list which files were touched and how many lines were changed in each. Press Ctrl+Shift+R to open the multi-buffer review page, allowing you to accept or reject changes section by section.

My advice is direct: always review the diff before accepting. The most common mistake an AI makes isn't flawed logic; it's "helpfully" modifying something you never asked it to change.

Advanced Tips

Open multiple threads to work in parallel. Each thread has its own independent context and history; use Cmd+Alt+J to open the thread sidebar and switch between them. If two threads might touch the same batch of files, use worktrees to isolate them and prevent conflicts.

Manage tool permissions with Agent Profiles. Profiles determine which built-in tools and MCP servers a thread can access; Tool Permissions decide whether operations requiring authorization are allowed, denied, or prompt you every time. When running on production projects, keeping "require confirmation" enabled is much safer.

Connect MCP servers to expand capabilities. The Zed Agent supports MCP, allowing you to hook up capabilities like database queries and documentation retrieval so the AI can see more than just code. To understand the value of MCP, check out our AI Coding Assistant Guide.

Turning off the AI is also an option. Setting "show_edit_predictions": false disables it globally, or you can turn it off for specific languages or target specific directories using disabled_globs. This is super handy when working on sensitive projects.

Things to Watch Out For

The plugin ecosystem is the biggest gap. Zed's extension mechanism is improving, but compared to VS Code's tens of thousands of plugins, it still has a long way to go. Before packing up and moving over, list the plugins you actually use every day and check for alternatives one by one.

Team inertia is harder to cross than technology. If you use Zed while your colleagues use VS Code, friction will arise during pair programming or shared debugging setups. The pragmatic approach is to start by using it on personal projects.

Inspect your project contents before launching the Agent. Letting an AI read and write across an entire project is equivalent to letting the model see your .env files, API keys, and client data. This applies regardless of the editor you use, but every time you switch tools is a great chance to double-check—especially since many freelance project repos in Taiwan actually contain client production environment keys.

Chinese input methods occasionally misbehave. Input method integration in certain scenarios still has minor hiccups, so Zhuyin/Bopomofo users may want to test it out firsthand before making a final call.

TheAI Academy Review

I've been using Zed for a while now, and what stands out the most isn't the AI—it's the "click and it's instantly open" lack of latency. Everyone has AI features nowadays, making speed the actual differentiator.

Review: Zed is "an editor built right first, with AI dropped in afterward," rather than the other way around. That sequence dictates the ceiling of its user experience.

Concrete advice for readers: Spend two weeks running personal projects on the free version with your own API key. This combination costs zero in subscription fees while letting you fully experience the Agent workflow. Two weeks later, if you find you can't go back to VS Code's sluggishness, consider paying for Pro; if it's missing too many plugins, you've only spent two weeks testing it with zero loss.

To compare other options, check out the Cursor Tutorial and GitHub Copilot Tutorial, or visit the Prompt Template Library to find ready-to-use development commands.

Sources

(This article is compiled based on public information and official documentation. Features and pricing are subject to the latest official announcements.)