Zed said on May 7, 2025, that it now offers what it called the world’s fastest AI code editor, putting its new Agent Panel at the centre of a broader push to combine speed, openness and agentic editing. The company presented the release as an extension of the editor’s existing design rather than a separate AI layer added on top of a conventional code tool.
The editor itself remains open source under GPLv3, and Zed said the new AI features are open source as well. That matters because the company’s pitch is not only that the editor is fast, but that users can inspect how the AI functionality works and decide how much access to give it. Zed said the Agent Panel can answer questions about a codebase, make changes directly and write new code when asked to do so.
The company described the system as a way to reduce the amount of setup work usually required before a model can be useful inside a large project. In Zed’s example, the agent was given a simple editing task in a blog post format and immediately began searching the codebase on its own to determine what needed to change. Zed framed that as a normal part of collaborative software work: when a developer enters an unfamiliar repository, the first step is often to orient themselves by searching and tracing the structure of the code.
The release also emphasises privacy and control. Zed said conversations with the agent are private by default and are not harvested for training unless a user explicitly opts in through feedback buttons. The agent also asks for confirmation before actions that could not easily be undone, such as running terminal commands, and users can set a preference so the system always asks first. According to the post, the agent is intended to work in the background without interrupting the user unless it needs attention.
Zed said the panel can access a unified diff after it completes a task, allowing the user to review and edit the results. The company also said the agent can make use of language servers, linters, formatters and terminal commands, subject to permission settings. Extensions can expand those capabilities further, and Zed said the product supports the Model Context Protocol, which can connect the editor to outside tools such as databases, analytics systems and browser automation.
The company tied the launch to a pricing model that still allows non-AI use without charge. Zed said users can keep using the editor as a standard non-AI tool on macOS and Linux, and that Windows support remains in progress. It also said users can bring their own API keys or run custom models through Ollama on their own hardware. On the paid side, Zed described a free tier with 50 prompts per month and a Pro tier with 500 prompts for $20 a month.
Zed’s message is that AI should feel native to the editor rather than bolted on as a separate product. The company said its long-term goal is to build a sustainable business mostly from optional paid features, while keeping the core editor available as a free and open-source tool. For developers deciding whether to try it, the company’s central claim is simple: the new Agent Panel is meant to make AI assistance fast, inspectable and optional at the same time.
