Real tasks on your real prototypes
Share one link. Participants join with a code, an email address or anonymously, on a phone or a laptop, with no account and nothing to install. Every task records time, success, first click and click path.
New in v0.3
Tree tests, card sorts and first-click heatmaps
Know what your users actually do.
UX Testbench is an open-source tool for surveys, A/B prototype tests, card sorts, tree tests and first-click tests. Self-host it, and get results that are honest about what the numbers can and cannot tell you.
No credit card. MIT licensed. Self-host it and we never see your data.
A verdict that can say “not yet”
Most tools run a t-test on eight people and print a winner. UX Testbench uses an exact sign test and bootstrapped intervals, and every verdict shows its p-value and how many paired participants it rests on.
- Evidence for the challenger
- Not enough evidence yet
- Evidence against the challenger
See where people click first
First-click tests turn every click into a heatmap over your screenshot, split by desktop and mobile. Below five participants it shows each click on its own, because a density map from so few would invent a pattern.
Build in the browser, or in YAML
Chain surveys, tests and screeners into one study in the Studio, or edit the YAML and keep it in version control. A scratch copy is validated before anything goes live, so a typo cannot take a running study down.
Five methods, one participant session. No plugins, nothing to install.
Surveys and screeners
Likert scales, multiple choice, matrices and open text, with conditional skip logic.
A/B prototype tests
Serve two HTML prototypes and record time on task, success, first click and click path.
Card sorting
Open, closed or hybrid sorts, with similarity matrices and dendrograms.
Tree testing
Text-only hierarchies measuring directness, success and backtracking.
First-click tests
Screenshots with click heatmaps and first-click success, by device.
…and any of them chained
Prerequisites and audience rules decide who sees which module, and when. One study instead of four.
Your research belongs to you.
No participant panel resold, no model trained on your data, and no analytics script following the people who help you.
- One database per studyArchiving a study is copying a file; deleting one leaves nothing behind.
- No third-party requestsWith LOCAL_ASSETS=1, participant pages load nothing from a CDN.
- MIT licensedSelf-host it forever, for free. No seat count, no feature held back.
self-host in a minute
# clone, install, run
$ git clone https://github.com/syaifullah18/ux-testbench
$ cd ux-testbench && pip install -r requirements.txt
$ python -m testbench run
* Running on http://127.0.0.1:5000
* No database server, no container, no build stepIs UX Testbench really free?
Yes. It is MIT licensed, so you can self-host it forever at no cost. The hosted version is free while it is in beta.
Can I use it for commercial research?
Yes. The MIT licence allows commercial use, including running studies for clients.
What do participants need?
Only the study link. It works on a phone, needs no account and no install, and participants can stop and resume later.
Where is the data stored?
On your own server. Every study gets its own SQLite file, and with LOCAL_ASSETS=1 participant pages contact no third party at all.
How many participants do I need?
Fewer than you might think for a direction, more than most tools admit for a decision. The report says "not enough evidence yet" until the sign test has enough paired participants to back a verdict.
How can I contribute?
Open an issue or a pull request on GitHub. CONTRIBUTING.md in the repository explains how to run the tests and what a good change looks like.
Start your first study today.
Free while it is in beta, and free forever if you host it yourself.