Run once and reuse?

[Explosion-Scratch]

The most useful application of Skyvern imo would be to have it run once (or once in a while) then reuse the workflow it generates. Could it be made to generate puppeteer code? Or some sort of YAML file that doesn't need AI to run again?

This would help significantly to cut costs and have more reliability while still benefiting from the AI integration

[cdelgadob]

That my opinion as well. I was thinking about the AI functionality as kind of a fallback mechanism. So it would be: run once and validate. Then run and if it fails use the AI again to adapt to possible changes. Or you could check if the page has changed before running the agent using some sort of cache. What do you think? Enviado desde Outlook para iOS<https://aka.ms/o0ukef>

…

________________________________ De: --Explosion-- ***@***.***> Enviado: Saturday, October 26, 2024 7:52:21 PM Para: Skyvern-AI/skyvern ***@***.***> Cc: Subscribed ***@***.***> Asunto: [Skyvern-AI/skyvern] Run once and reuse? (Issue #1062) The most useful application of Skyvern imo would be to have it run once (or once in a while) then reuse the workflow it generates. Could it be made to generate puppeteer code? Or some sort of YAML file that doesn't need AI to run again? This would help significantly to cut costs and have more reliability while still benefiting from the AI integration — Reply to this email directly, view it on GitHub<#1062>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABCPTEQHQYSUDPFJZAJFKELZ5PJFLAVCNFSM6AAAAABQU76ZAOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGYYTMMBTGM4DSOA>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[suchintan]

We have this in beta right now. If you create a task block inside workflows, there's a "cache" option under advanced

Cached actions don't use LLMs under the hood!

[cdelgadob]

That’s good news! And what happens if the task fails? Does it fallback to LLM to retry? Enviado desde Outlook para iOS<https://aka.ms/o0ukef>

…

________________________________ De: Suchintan ***@***.***> Enviado: Saturday, October 26, 2024 10:54:54 PM Para: Skyvern-AI/skyvern ***@***.***> Cc: Carlos Delgado ***@***.***>; Comment ***@***.***> Asunto: Re: [Skyvern-AI/skyvern] Run once and reuse? (Issue #1062) We have this in beta right now. If you create a task block inside workflows, there's a "cache" option under advanced Cached actions don't use LLMs under the hood! — Reply to this email directly, view it on GitHub<#1062 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABCPTEWKRRAPXI5VPEYENA3Z5P6R5AVCNFSM6AAAAABQU76ZAOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMZZG4ZTGNJTGA>. You are receiving this because you commented.Message ID: ***@***.***>

[suchintan]

Yeah exactly. If the cache fails, it falls back to the LLM and adds the successful thing to the cache