Wu: a native Rust editor with coding agents built in
Wu is a native Rust code editor with Claude Code, Codex and OpenCode built in, plus a published Linux VM memory test against Zed and VS Code.
Project facts
GitHub Ecosystem- Language
- Rust
- Stars
- 647
- Data checked
- 2026-11-12
Snapshot figures reflect the check date and may change over time.
If you want Claude Code, Codex, or OpenCode available inside your editor, Wu is a native code editor written in Rust and an agent workspace in the same app. It began as a Zed fork, though the project says it has diverged substantially. GitHub showed 647 stars on November 12, 2026. The useful distinction is the combination of built-in coding agents and a public memory comparison with enough detail to judge its limits.
Core features
- Coding agents in the editor: The README lists Claude Code, Codex, and OpenCode. Developers who normally launch those tools from a terminal can work with them alongside their code. Wu is not described as a hosted model service, and the README does not claim that a cloud account is required.
- A native Rust desktop app: The project says it uses neither Electron nor webviews and offers builds for macOS, Windows, and Linux. That describes its implementation; it does not guarantee a faster or smaller result for every project or machine.
- Packages across three operating systems: The README lists macOS builds for Apple Silicon and Intel, Windows x86-64, and Linux x86-64 and AArch64. macOS and Linux can use the install script. On Linux, the script places the command under the user’s local directory, so
~/.local/binmay need to be added toPATH. - An account and telemetry claim: Wu’s README says there is no account requirement and no usage telemetry. Treat that as the project’s statement, not as an independent network audit.
- A published editor memory test: Wu compares process-tree PSS for idle screens and workspace tasks such as opening files, searching a repository, and generating terminal output. The numbers below come from a software-rendered Ubuntu VM, with the setup described alongside them.
Typical use cases
- Rust developers who want to try a Zed-derived editor while keeping Claude Code, Codex, and OpenCode available from one workspace.
- Developers moving between macOS, Windows, and Linux who need a native desktop editor. The listed downloads cover those systems, but users should match the package to their CPU architecture.
- Teams comparing editor memory use. The project publishes a Linux ARM64 setup that can be reproduced before repeating the comparison with a team’s own repositories, language servers, and hardware.
- People running coding agents in parallel. Worktrunk can manage separate Git worktrees for those tasks; it addresses workspace isolation, not editor performance.
Quick start
On macOS or Linux, the README’s install command is:
curl -fsSL https://wu.dstrml.org/install.sh | sh
Windows users can download the x86-64 installer from GitHub Releases; macOS also has direct DMG downloads. The script handles macOS and Linux, and Linux installs wu under ~/.local/bin. If the shell cannot find the command, add that directory to PATH as the installer explains, then launch Wu from the application menu or terminal.
Once the editor opens, connect Claude Code, Codex, or OpenCode through the built-in agent entry points. The README does not reduce setup to one universal command, so the exact connection flow depends on the current app version and its documentation. Open a repository you know, check that its language tools work, and confirm each agent separately before moving a primary workflow into Wu.
Benchmarks
The project’s benchmark document compares Wu 1.0.10, Zed 1.21.0, and VS Code 1.139.1 using their official Linux ARM64 releases. It reports process-tree PSS in MiB; lower is less memory for that measured stage.
| Scenario | VS Code | Zed | Wu | Wu below VS Code | Wu below Zed |
|---|---|---|---|---|---|
| Idle welcome window | 743.5 | 420.4 | 382.5 | 48.5% | 9.0% |
| Repository open, no source file | 725.2 | 423.7 | 381.2 | 47.4% | 10.0% |
| Open 20 source files | 826.2 | 554.3 | 516.9 | 37.4% | 6.7% |
| Search the repository | 823.6 | 1,395.3 | 582.9 | 29.2% | 58.2% |
| Generate terminal output | 1,405.5 | 1,664.4 | 862.8 | 38.6% | 48.2% |
The table omits three workflow stages: edit, save and scroll (35.0% below VS Code), navigate and search a large log (37.2%), and close tabs (14.5%, the smallest gap to VS Code). The full set is in the project’s memory-benchmark.md.
The host was an Ubuntu 26.04 ARM64 VM with six Neoverse-N1 vCPUs, 7.7 GiB RAM, a 1280 × 800 virtual display, and no GPU. Xvfb and Mesa llvmpipe supplied CPU software rendering. The active workflow used a Wu source snapshot, disabled language servers, and included edit/save cycles, searches, a 21.9 MiB log, and terminal output. The project says each figure is the median of settled-memory medians from three independent runs. Software-rendering allocations are part of the result, so it is not a prediction for a hardware-GPU desktop. Operating systems, dependency graphs, debugging, and signed-in AI features can change memory use.
Summary
Wu is worth a look for developers who want a native Rust editor with several coding agents in the same workspace. Its memory table is more useful than a bare “lightweight” claim because it names versions, VM resources, and a measurement method. It remains one project-run environment, not evidence that Wu will use less memory than Zed or VS Code on every desktop. GitHub reports NOASSERTION for the repository’s license field; the README says Wu is GPL-3.0-or-later with Apache-2.0 components where marked. Read the relevant license files before using or redistributing it. Rust is the primary language. This entry relies on the README and docs; the tool was not installed or run here.