How to use Zed: the fast, open-source editor with AI

Zed Industries' open-source code editor Zed, written from scratch in Rust and licensed primarily under GPL-3.0-or-later, ships an Agent Panel that lets developers run Claude, Codex, or Gemini CLI inside the editor and connect models via their own API key, Zed-hosted models billed at the provider's API list price plus 10%, or a free local model through Ollama. Stable builds are available for macOS, Linux, and Windows via `brew install --cask zed`, `winget install -e --id ZedIndustries.Zed`, and `curl -f https://zed.dev/install.sh | sh`, with no web version released yet. The editor is free, and the agent panel opens with Cmd+Shift+A on macOS or Ctrl+Shift+A on Linux and Windows.

The Primer · Tools & Stacks How to use Zed: the fast, open-source editor with AI Vendor-neutral Zed is a code editor written from scratch in Rust for speed, open source, and free. Its AI is what makes it a vibe-coder tool: an agent panel that edits your code, a one-line path to a free local model, and the ability to run Claude, Codex, or Gemini CLI inside the editor over an open protocol. Here is the whole thing, install to agent. Zed is the editor for people who want speed and want to own their AI stack. It is written from scratch in Rust by the team behind Atom and Tree-sitter, which is why it feels faster than the Electron editors most of us grew up on, and it is open source and free. What makes it a vibe-coder tool rather than just a fast text editor is the AI: an agent panel that edits your code, a one-line path to a free local model, and the ability to run Claude, Codex, or Gemini CLI inside the editor. For a builder who wants a quick editor that does not lock them to one model vendor, that is the pitch. Here is the whole answer up front. Install Zed, open the agent panel with a keystroke, and point it at a model: your own API key, Zed’s hosted models, or a local one running free through Ollama. If your agent of choice is Claude Code or Codex, Zed runs it inside the editor over an open protocol. The editor itself costs nothing, and everything below is the how. What Zed is Zed bills itself as a high-performance, multiplayer code editor, and the Rust-and-GPU foundation is the reason the speed claim holds up in practice. It is open source, licensed primarily under GPL-3.0-or-later, built by Zed Industries. The parts that matter for this guide are that it natively supports agentic editing and does not tie you to a single model provider, which is exactly the model-agnostic camp our agent field guide https://okaneland.com/study/ai-coding-agent-field-guide/ places it in. Install it on macOS, Linux, or Windows All three desktop platforms are available as stable builds a web version is not out yet . One command each, from the install docs read 15 August 2026: - macOS: brew install --cask zed - Windows: winget install -e --id ZedIndustries.Zed - Linux: curl -f https://zed.dev/install.sh | sh Or download directly from zed.dev. Windows searchers ask about this a lot, so to be clear: the Windows build is a stable download alongside macOS and Linux, installable through winget. The AI agent panel Zed’s AI lives in the Agent Panel, which the docs describe as where you interact with agents that can read, write, and run code in your project. Open it with Cmd+Shift+A on macOS or Ctrl+Shift+A on Linux and Windows, or run agent: new thread from the command palette, or click the sparkle icon in the status bar. That panel is the surface for generating code, refactoring, and asking questions about your project, and everything in the next sections is about what model powers it. Pick your model: your key, Zed-hosted, or local Zed gives the agent five ways to reach a model, and you will use one of three. Bring your own API key for a provider Anthropic, OpenAI, Google AI, DeepSeek, Mistral, GitHub Copilot, LM Studio, or Ollama , which you save in the panel’s settings and which then costs only what that provider charges. Use Zed’s hosted models on a paid plan, where Zed runs major models for you and bills hosted usage at the provider’s API list price plus 10%. Or run a local model , which is the free path and the next section. OpenRouter, Vercel, and Amazon Bedrock are wired in as gateways rather than direct providers, if you route through one of those. Run a local model with Ollama This is the most-searched thing about Zed’s AI, and the setup is short. Install Ollama, pull a model with ollama pull mistral or a coding model , and run ollama serve , or just open the Ollama app on macOS. Zed treats Ollama as a dedicated first-party provider, so the model appears in the agent panel’s model picker once the server is up. If you need to point Zed at a non-default address or set the context window, it lives under the language models key in settings.json : { "language models": { "ollama": { "api url": "http://localhost:11434" } } } The context length defaults to 4096 tokens, which you can raise there too. That is a coding agent running against a model on your own machine for no per-token cost, the same zero-cost setup our Ollama guide https://okaneland.com/primer/how-to-use-ollama/ covers from the model side; LM Studio https://okaneland.com/primer/how-to-use-lm-studio/ is a supported provider here as well if you prefer its window. Run Claude, Codex, or Gemini CLI inside Zed This is the feature worth switching for, and it is the most on-brand for a vibe coder. Zed connects external agents through the Agent Client Protocol ACP : the external agent keeps its own runtime and login, while Zed hosts the conversation in its agent panel. The documented agents include Claude as the Claude Agent , Codex, OpenCode, Copilot, Cursor, and Gemini CLI. So if you already run Claude Code, you keep it, its subscription, and its behavior, and you drive it from inside Zed instead of a separate terminal, with your editor’s context right there. It turns Zed into a neutral home for whichever agent you already pay for, rather than a competitor to it, which is a genuinely different proposition from an editor that ships one built-in assistant. Edit predictions with Zeta Separate from the agent panel is inline autocomplete, which Zed calls edit predictions. The default model is Zeta, an open-source model Zed developed, and you accept a suggestion with tab. You can swap in GitHub Copilot or Mercury Coder if you prefer. The free tier includes 2,000 accepted Zeta predictions a month, and the Pro plan removes that limit, so most people never think about it until they are coding heavily. What it costs, and the open-source part The editor is free and open source, and the pricing is only about the AI. The free tier is called Personal and is $0 forever: it includes those 2,000 monthly edit predictions and, importantly, unlimited use of the agent panel with your own API keys or with external agents like the Claude Agent and Codex CLI. In other words, bring your own key or your own Claude Code subscription and the whole thing costs nothing beyond what you already pay that provider. Pro is $10 a month for unlimited edit predictions plus Zed’s hosted models with $5 of included tokens, and Business is $30 a seat for team governance. Add the GPL-3.0 license and the free local-model path, and Zed is one of the few editors where a cost-conscious builder can run a full agentic setup for genuinely $0, which is the same calculation our cost study https://okaneland.com/study/cut-your-ai-coding-bill/ keeps returning to. Install it, open the agent panel, point it at a local model or the agent you already run, and you will know within an afternoon whether the Rust-native speed and the bring-your-own-everything model fit how you work. One email, when there's something worth sending Get the research in your inbox. No fixed schedule, no filler. You get an email when we've tested something, run the numbers, or found a tool worth your time. Free. Double opt-in, unsubscribe in one click. What did you plug into Zed? Get the research → https://okaneland.com/subscribe/ Sources | Source | Link | |---|---| | Zed README and homepage github.com/zed-industries/zed and zed.dev, read 2026-08-15 : Zed is a high-performance, multiplayer code editor from the creators of Atom and Tree-sitter, written from scratch in Rust to use multiple CPU cores and the GPU; it is open source, with built-in collaboration and AI, and natively supports agentic editing. Source is licensed primarily under GPL-3.0-or-later with Apache-2.0 components where marked. | github.com ↗ https://github.com/zed-industries/zed | | Zed install docs zed.dev/docs/installation and /docs/windows, read 2026-08-15 : available on macOS, Linux, and Windows all stable; web is not yet available . Install with brew install --cask zed on macOS, winget install -e --id ZedIndustries.Zed on Windows, or curl -f https://zed.dev/install.sh \| sh on Linux. | zed.dev ↗ https://zed.dev/docs/installation | | Zed AI docs: agent panel and configuration zed.dev/docs/ai/agent-panel, /ai/configuration, read 2026-08-15 : the Agent Panel is where you interact with AI agents that read, write, and run code in your project; open it with Cmd+Shift+A macOS or Ctrl+Shift+A Linux/Windows . Model access has five paths: Zed-hosted models, your own provider API key, an existing subscription, a gateway, or a local model. Bring-your-own providers include Anthropic, OpenAI, Google AI, DeepSeek, Mistral, GitHub Copilot, LM Studio, and Ollama; OpenRouter, Vercel, and Amazon Bedrock are treated as gateways. | zed.dev ↗ https://zed.dev/docs/ai/configuration | | Zed local-model docs zed.dev/docs/ai/use-a-local-model and /docs/configuring-zed, read 2026-08-15 : to run a local model via Ollama, install Ollama, ollama pull mistral, run ollama serve or open the Ollama app on macOS , then select the model in the agent panel. Ollama is a dedicated first-party provider configured under the language models key, e.g. {"language models": {"ollama": {"api url": "http://localhost:11434"}}}; the context length defaults to 4096 num ctx . | zed.dev ↗ https://zed.dev/docs/ai/use-a-local-model | | Zed external agents and edit predictions zed.dev/docs/ai/external-agents and /ai/edit-prediction, read 2026-08-15 : external agents integrate through the Agent Client Protocol ACP , where Zed hosts the thread in the Agent Panel while the external agent owns its runtime; documented agents include Claude as Claude Agent , Codex, OpenCode, Copilot, Cursor, and Gemini CLI. Edit predictions default to Zeta, an open-source model developed by Zed, accepted with tab, with Copilot and Mercury Coder as alternatives. | zed.dev ↗ https://zed.dev/docs/ai/external-agents | | Zed pricing zed.dev/pricing, read 2026-08-15 : Personal is $0 forever with 2,000 accepted edit predictions and unlimited use with your own API keys or external agents like Claude Agent and Codex CLI; Pro is $10/mo with unlimited edit predictions, $5 of included tokens, usage-based billing beyond that, and Zed-hosted models hosted usage bills at API list price plus 10% , with a two-week trial; Business is $30/seat/mo with org-wide model policies, governance, and role-based access. | zed.dev ↗ https://zed.dev/pricing |