On this page
A live test of synchronization work
The live benchmark starts three Yjs document replicas with the same text. Each replica makes a small edit. It compares sending incremental document updates with re-sending whole-file state, then checks whether all replicas agree. The page reports measured timing and update sizes after warm-up.
The benchmark runs in a Web Worker. It repeats merge operations to make short durations readable on browser clocks. Results depend on your device, browser, workload and background activity; the live page explains its sampling and averaging.
What the numbers tell you
- Merge duration measures local work applying updates, without network latency.
- Update size measures encoded data for this workload, not the bandwidth of a complete collaboration session.
- Replica agreement checks that the copies have the same text. It does not prove that the code compiles or meets its requirements.
- The full-file baseline is a synchronization strategy, not a measured version of another editor.
How to reproduce and share a measurement
- Open the live test on the device and browser you want to evaluate. Let the warm-up complete.
- Record the browser and operating system versions, hardware, workload, date, sample count and whether other applications were busy.
- Record both timing and update size, together with any divergence or clock-resolution warning shown by the page.
- Repeat under the same conditions. Keep the individual results rather than choosing only the fastest run.
A session across two machines needs separate network measurements. Record round-trip delay, connection failures and reconnection time alongside local merge work. One fast local test cannot establish a peer-to-peer latency promise.
Evaluating the desktop app
The shipping desktop app uses Tauri with a web frontend and Rust commands for native operations such as filesystem access and terminals. It still uses a system webview. This architecture alone does not establish a particular idle memory footprint, frame rate or input latency.
For a useful comparison, use the same repository, files, language services, extensions and agent workload in each editor. Separate cold launch from warm launch, record peak and steady memory, and publish the procedure and raw samples before making numerical claims. Until then, Crux makes no fixed hardware performance promise here.
See Tauri's architecture overview for the distinction between the frontend and native runtime, or compare workflows to decide which editor fits your team.
BUILD SOMETHING TOGETHER
Your team.
Your agents.
One workspace.
Bring your people, code and context together in Crux.
Join the waitlist