Back to catalog
Zed
GPL-3.0Zed is an open-source, GPU-accelerated code editor built in Rust that offers local LLM privacy, collaborative editing, and minimal memory consumption.
Overview
Security-conscious teams working in regulated environments often struggle when default IDE settings transmit file contents and usage metrics to third-party model providers. Zed counters that friction by running as a compiled binary with local configuration files and native offline workflows. Developers retain explicit control over outbound network traffic, language server binaries, and artificial intelligence backends.
Where Zed fits
Most modern code editors rely on web runtimes and cloud-first telemetry pipelines. Electron-based workstations like Visual Studio Code and Cursor bundle extensive JavaScript layers, run background update daemons, and route automated model interactions through hosted vendor gateways. Zed takes the architectural path of Sublime Text instead: written in Rust, it renders directly via the GPU using a native UI framework and bypasses the Chromium abstraction layer. It provides instant startup, lower resident memory consumption, and predictable responsiveness when navigating large repositories or editing log files.
From a governance perspective, Zed separates local editing from network dependencies. While hosted editors nudge users toward proprietary account logins and telemetry tracking, Zed functions entirely offline with simple JSON configuration files. Teams can disable client diagnostics with a single toggle, point syntax engines to system-installed binaries, and route AI requests directly to self-hosted Ollama servers or internal OpenAI-compatible gateways. The tradeoff is ecosystem depth: Zed relies on a sandboxed WebAssembly extension model, meaning developers cannot install the vast library of Node.js-based Visual Studio Code extensions or arbitrary embedded webviews.
What it does well
- GPU-accelerated rendering draws text at display refresh rates, eliminating input latency when scrolling large codebases or configuration trees.
- Tree-sitter integration parses syntax directly into concrete syntax trees on background threads, maintaining accurate token highlighting during rapid keystrokes.
- Flexible model routing connects the assistant panel and inline transformation tools to self-hosted Ollama instances, Anthropic, or OpenAI endpoints without routing code through third-party proxies.
- Multi-buffer editing lets developers open multiple search results or symbol references across different files in a unified virtual buffer, enabling batch refactoring without switching tabs.
