I’m using n8n for this purpose, running both a lot of automation as well as api for the weweb frontend.
While it is great it still comes in as a good bit slower than running close to db functions in postgres/supabase.
I found a certain completely no-code workflow in n8n was 1/10 the speed of the same function running as a postgres function. I was able to get it down to 300 ms (compared to 100-150 ms on postgres) by using the code nodes in n8n, instead of if/switch and other dedicated nodes - and then it’s not really no code any more.
Running n8n as close to db as possible is important, and queue mode is also a must to keep a free non-blocking environment for frontend api calls. I run main+webhook node and 4 workers now.
Yeah, I’d say with the current state of the AI , you can learn a lot while needing very few to get started so I feel like n8n’s cons are just too painful. A coded Edge Function is the way. You can even make one function into multiple endpoints by using something like Oak for routes. So you are not limited by the amount of Edge Functions. E.g you can have an /auth function that has /login, /signup, /special-signup etc. All in one Edge Function, which is very convenient.
I’m not saying you’re wrong, I’m just asking for optimum visual development options for low latency, and still offering the flexibilty of visual development.