Performance
Ironcall vs the field
Raw numbers comparing startup time, memory usage and binary size across the major HTTP API clients.
How to read these numbers
Measured on a single Linux machine (specs below), median of 5 runs, all clients forced onto the same windowing backend. Absolute values scale with your hardware; the relative gap between clients is the meaningful signal.
| Metric | Ironcall← ours | Postman | Insomnia |
|---|---|---|---|
Cold startup Launch to visible window, OS cache dropped first | ~95 ms | ~470 ms | ~1 080 ms |
Warm startup Second launch, OS cache warm | ~42 ms | ~256 ms | ~568 ms |
RAM idle Empty workspace, full process tree (PSS) | ~280 MB | ~660 MB | ~820 MB |
Install size Installed package on disk | ~29 MB | ~395 MB | ~1 538 MB |
* Measured 2026-06-11 on the machine described below. See methodology.
Methodology
Test environment
- Hardware: Intel Core i9-14900KF, 31 GB RAM
- OS: EndeavourOS, Linux 7.0.10-zen1
- Versions: Ironcall 0.8.1, Postman 12.12.6, Insomnia 12.6.0
Conditions
- Each tool measured in isolation, clean session
- Cold start: process killed + OS cache dropped between runs
- Startup = time to visible window, not time to interactive
- RAM = PSS of the full process tree
- Median of 5 runs, all clients on XWayland
A caveat on startup times
The startup figures measure time to the first visible window, not time to a fully interactive UI. Tauri maps its native window early, while Electron shows its window after Chromium has booted, so this metric favours the native-webview approach. We think that reflects a real difference (no bundled browser to start), but it is not the same as time-to-interactive. A proper time-to-interactive comparison needs more work, and we will publish it once we have a measurement we trust.
Why does this matter?
Most API clients are built on Electron, which bundles a full Chromium browser and a Node.js runtime. Ironcall is built on Tauri, which uses the OS's native webview and a Rust backend. The result is a dramatically smaller binary, lower idle memory, and a startup time measured in milliseconds rather than seconds.
Reproduce these results
The methodology and machine above are everything you need to reproduce these on your own hardware. If you run your own tests and get different numbers, let us know.
See the difference yourself: download Ironcall and compare.
Download Free