I'm Ignacio, CTO at Timeline. I launched an OSS library named Sedum today that lets you write e2e tests in plain english (or any other language).
Right now we are shipping so much AI generated code that we needed a way to write AI e2e tests in an economical way that we can run on every PR or even multiple times in local environments. The existing platforms charge you by step and they add a lot of mark up to the underlying AI calls.
For our 20 person engineering team, running these AI e2e tests in one of the existing paid tools, meant having to spend more than $5k/mo.
I was already thinking of doing an Open Source alternative some time ago, and when Jev launched, I played with it and quickly realised it's the way to go. It really makes it economical to run these tests with very high frequency and its probabilistic nature also helps you flag potential flakiness upfront, understand which tests should be run depending on the files being changed, and many other interesting use-cases.
But.. because Jev is only text and sometimes pure text is not enough to translate a sentence from plain english into a proper action or a proper assertion, in the cases where Jev is not confident enough, our library falls back on a vision model such Gemini Flash 3.8 (again.. BYOK).
Anyway, check it out and let me know what you think :)
Hey HN,
I'm Ignacio, CTO at Timeline. I launched an OSS library named Sedum today that lets you write e2e tests in plain english (or any other language).
Right now we are shipping so much AI generated code that we needed a way to write AI e2e tests in an economical way that we can run on every PR or even multiple times in local environments. The existing platforms charge you by step and they add a lot of mark up to the underlying AI calls.
For our 20 person engineering team, running these AI e2e tests in one of the existing paid tools, meant having to spend more than $5k/mo.
I was already thinking of doing an Open Source alternative some time ago, and when Jev launched, I played with it and quickly realised it's the way to go. It really makes it economical to run these tests with very high frequency and its probabilistic nature also helps you flag potential flakiness upfront, understand which tests should be run depending on the files being changed, and many other interesting use-cases.
But.. because Jev is only text and sometimes pure text is not enough to translate a sentence from plain english into a proper action or a proper assertion, in the cases where Jev is not confident enough, our library falls back on a vision model such Gemini Flash 3.8 (again.. BYOK).
Anyway, check it out and let me know what you think :)