Zed

An open-source code editor with an agent panel, edit predictions, and support for external agents – bring your own API keys or use local models.

Zed is a code editor that does not bolt AI on as a plugin but treats it as part of the workspace from the ground up. It is built by Zed Industries, the team behind Atom and Tree-sitter; the repository describes Zed as a high-performance editor with multiplayer features. Anyone looking at Zed as an AI assistant for programming therefore does not get a chatbot in the browser but a complete editor for macOS, Linux, and Windows in which agents, suggestions, and chat work right next to your own code.

The AI features Zed ships with

According to the provider's documentation there are four building blocks. The agent panel is where you give agents tasks, add context, review proposed changes, and manage several running conversations. Edit predictions suggest the next change while you type, and a single keystroke accepts it. The inline assistant transforms a selected piece of text in place after a short instruction. And through external agents you can plug in further tools that run in their own process with their own configuration and talk to the editor via the Agent Client Protocol (ACP). That is a clear difference from editors that hard-wire exactly one agent.

Your own models, your own keys, local execution

Zed is not tied to a single model provider. According to the provider, you can use Zed-hosted models, your own API keys, supported subscriptions, gateways, and local models. If you store your own key with a model provider, you are billed there directly; if you run a local model on your own server, your requests stay entirely inside your network. External agents such as Claude or Codex CLI can, per the pricing page, be used without additional charges on Zed's side. For teams that want to control model choice and data flow themselves, this is an important point.

MCP and extensibility

The Zed agent can pull in additional tools through the Model Context Protocol. This lets you use servers from the Skill Road catalog – for databases, browser automation, or cloud services, for example – in the agent panel without switching editors. Together with the built-in tools and skills, you get a workspace that adapts to your project instead of dictating a fixed feature set.

Privacy and security

The documentation is open about which data goes where: depending on the path you choose, prompts and code context are sent to Zed-hosted models, to your own provider, to a local server, or to an external agent. With edit predictions, local editing context can be sent to the provider. According to the provider, the agreements with model providers prohibit training on prompts and code, and zero-data-retention commitments exist for Zed-hosted models; for individual models designated by the model provider, a time-limited retention for safety review applies that Zed cannot switch off. Data is only kept if you actively take part, for instance with ratings or by enrolling in the training-data collection for edit predictions in open-source projects. The details are on the provider's privacy page and are worth reading before using the editor with sensitive code.

License, pricing, and plans

According to the repository, the source code is mostly licensed under GPL-3.0-or-later, with some components under Apache-2.0. The editor itself is free. For the AI features, the provider lists three plans: a permanently free Personal plan with a limited number of accepted edit predictions and unlimited use with your own API keys or external agents; a Pro plan with unlimited edit predictions, access to Zed-hosted models, and a monthly token credit included; and a Business plan with organization-wide model policies, data controls, role-based permissions, and a spend overview. Concrete amounts change and are listed on the official pricing page; Skill Road deliberately names no figures here.

Who Zed fits

Zed suits developers who want a fast, lean editor and want to keep control over which model works with their code. If you already use Claude, Codex, or another agent in the terminal, you can keep using it inside the editor instead of relearning a workflow. Teams benefit from the multiplayer features and the policies in the Business plan. If you would rather have a fully managed all-in-one tool with a fixed agent, the catalog offers alternatives such as Cursor or Kiro. A sensible way to start is to install Zed, connect the free plan or your own key, and let the agent panel handle one small, clearly scoped task in a test project.

Free
Provider
Zed Industries
License
GPL-3.0-or-later
Last reviewed
11.10.2026

Categories