Cache OpenAPI responses for the repeated regression test to speed up tests and reduce costs

[avkonst]

Hi, You project is an interesting idea. It would be good to have steps automatically cached and rerun from cache. If a step is failed, try running without cache and re-cache.

[seyacat]

I made a fork implementing cache. It stores the steps performed by the AI ​​and then reproduces them in each test. It may be of interest to you https://github.com/seyacat/auto-playwright

[earonesty]

fork is perfect. what would even be cooler is if you could "compile" the tests.

• run tests locally (hits openai)
• cache tests locally (openai cache is saved into a local folder)
• commit cache to github
• when tests run in ci, they don't have to hit openai... everything is cached

[bpetty-interlace]

I was just about to recommend this, and saw this request. @seyacat, great work. AI is only needed the first time it runs, ideally, and when it breaks. Huge cost savings. I honestly think that is why ZeroStep is so expensive, they are mostly just passing the cost of running the AI model on to the customer.

[bpetty-interlace]

@seyacat, would it be better to issue a PR to merge your fork into this repo since this functionality can GREATLY reduce cost and increase speed? It seems like this repo is actively maintained.

[seyacat]

I merge main in my fork, and add Deepseek support. works nice I think