Zed

The Zed AI code editor is an open‑source IDE built by the former Atom team. It uses Rust and the GPU to keep typing latency in milliseconds even inside million‑line repositories. On top of that speed, Zed adds a Zed AI agent panel, real‑time voice‑based pair programming, and native Git tools, so you can chat with LLMs, co‑edit live, and merge code without ever leaving the window.

This article explains what Zed offers today, walks through its key features, shows who’s using it, and outlines the qualities that set it apart from other AI‑assisted editors.

What is the Zed AI Code Editor?

Zed is a next‑generation source‑code editor available today on macOS and Linux, with a private Windows beta and a stable release planned for later in 2025. The editor itself is GPL‑licensed, while its GPU‑driven UI framework GPUI ships under Apache 2, so security teams can audit, fork, and extend without legal friction.

Beyond raw editing speed, Zed ships a full AI layer. You can pick Claude, Gemini, or any local Ollama model, and type natural‑language requests such as “add logging to this function” or “explain this SQL query.” API keys live in your local config by default, or you can purchase hosted usage through Zed’s plans. This gives teams precise cost control and compliance‑friendly Zed AI integration options.

Key Features

• Agentic Editing – Simply start a new thread and describe the change. Zed AI agent will patch the file, show a diff, and ask for your approval.
• Zeta Edit Prediction (AI Autocomplete) – The open‑source Zeta model predicts the next edit, not just the next token, so you can accept multi‑line changes with one tab. This Zed AI autocomplete runs locally, free while in beta, and includes 2000 predictions per month on the Personal plan.
• Inline Assistant – Highlight code, hit ⌥‑A, and the model refactors or documents the block in place, with no context switch.
• Built‑in Git and multi‑language debugger – Work end‑to‑end inside Zed: stage files, generate AI‑written commit messages, push to your remote. Supports debugging Rust, Go, Python, C/C++, and JavaScript through the new DAP‑based debugger without requiring any external plugins.
• Flexible, privacy‑first AI options – Use Zed’s hosted LLM quota (50 prompts on the free tier, 500 on Pro), swap in your own API keys for Claude or Gemini, or keep everything on‑prem by running local models with Ollama.

Who Is Using Zed?

Zed’s own site lists endorsements from well‑known engineers already using the editor in production:

• José Valim (Creator of Elixir) — “Zed has all of the features I could ask for in an editor.”
• Dan Abramov (React core team) — “This is obviously a product built with love and care. Kudos to the team.”
• Mike Bostock (Creator of D3.js) — “Thoughtful, precise design. And the speed, the speed!”

These testimonials signal that Zed’s performance, design, and AI tooling already meet the demands of serious, large‑scale workflows.

What Makes Zed Unique?

• Game‑engine performance – Built with Rust and its own GPU UI stack (GPUI), Zed renders screens at ≈120 FPS, turning searches and scrolls that take seconds in Electron‑based IDEs into millisecond tasks.
• Open‑by‑default stack – The core editor is GPL, GPUI is Apache 2, and the new Zeta edit‑prediction model is fully open source and retrainable. So security teams can audit every line or fine‑tune the model on internal code.
• Local‑first AI design – AI runs inside the editor and can even stay on‑prem, keeping proprietary code off third‑party servers.
• Real‑time multiplayer – Voice chat + CRDT cursors let teams pair‑program or demo without screen‑sharing lag or merge headaches.
• Release velocity – In just one quarter, Zed shipped a DAP‑based cross‑language debugger, native Git with AI commit messages, the Zeta edit‑prediction engine, and a Windows private beta.

Measurements

Zed can look impressive almost immediately because the editor feels fast, the AI features stay close to the code, and collaboration does not require jumping between tools. That still does not answer the practical question: is the team actually shipping faster with less review drag, or just producing changes more quickly at the start. Milestone helps make that visible.

The measurements that usually matter are simple:

• Time from prompt or edit request to first reviewable change
• Review time on Zed-assisted pull requests
• Acceptance rate of AI-generated edits or completions
• Follow-up edits needed after the first suggested patch
• Rework caused by missed context in multi-file or agent-driven changes

Improvements

Once that pattern is visible, the next step is usually deciding where Zed should stay in the workflow and where it needs tighter boundaries. Milestone is useful here because it helps teams improve usage based on delivery results, not just the appeal of a faster editor with built-in AI.

A few improvement areas usually show up early:

• Keep Zed focused on bounded edits, refactors, and reviewable changes
• Use tighter prompts for agent-driven or inline assistant tasks
• Watch for repeated cleanup after multi-line predictions or patches
• Apply stricter review to core logic and multi-file edits
• Separate quick productivity wins from broader architectural work

In many teams, Zed works best when the task already has a clear shape. Small refactors, documentation updates, test support, and scoped code changes often benefit most, while broader design-heavy work still needs slower human judgment.

Pricing

Personal – Free forever

• 50 hosted prompts
• 2000 local edit predictions
• Unlimited open‑source usage

Pro – $20/seat/month

• 500 hosted prompts,
• Usage‑metered afterward, so GPT bills never spiral

Enterprise – Custom pricing

• SSO
• On‑prem or air‑gapped deployment
• Role‑based dashboards
• Policy‑driven integration limits for strict compliance environments

Conclusion

Zed combines GPU‑speed editing, live collaboration, and plug‑and‑play AI into one open‑source package. Regular feature updates and transparent costs make the Zed AI code editor a serious challenger to VS Code, JetBrains, and cloud‑first copilots. If you need lightning performance plus on‑prem AI agents without vendor lock‑in, Zed is a great option to check out.