What sure fire ways have you found success in Agent using Replit Object Storage successfully on the first few development cycles?
I really like the idea of using the ROS, it feels simpier to me than bringing in an AWS bucket (although to be fair the Agent tries to use AWS every time, so maybe it’s easier). I’ve always been able to get it to work but it always seems to have the same sequence of problems no matter what tricks I try from the forums! It goes
Agent thinks on what to do, then brings in AWS libraries
Agent asks for AWS credentials, even though in my initial prompt, I give the ROS bucket ID
Agent moves onto google cloud services libraries, with about the same sequence as above.
Three to six prompts later it reallizes what I want and uses the replit bucket, but never with full success.
Finally, it will think on the topic a bit (I’ve tried the extended thinking too) and it will come up with the solution(s) and function without a hitch.
It’s not a huge deal, but it is tiresome to have the same problems even when you try different ways to create the prompt.
Perhaps I am going off topic here, but what you are describing isn’t unique to buckets. For me: “don’t use drizzle” - and what is the first thing it does!
This is typical of the agent process for many things in the way it just ignores clear instructions. I hope Replit are working on this, and I am sure the new replit.md will help to cement our instructions:
But I’m homing in on a process - a bit like managing your junior dev and knowing his/her shortcomings, so finding a way to support them, rather than scream and eventually sack them
prompt 1 tells it what to build and the overall scaffolding architecture. give it the basics of your app idea, but tell it you will enhance the features later. The purpose and focus of this prompt is really the scaffolding, not your idea
prompt 2 asks it to analyse the code and create an architectural compliance audit against a checklist you give it
prompt 3-… tell it to fix the bits it had obviously ignored and got wrong, and to come back inline with your original architecture
only once the above is done and a re-audit is clean, would you then start building out the features of your actual app idea. Do not start that until you have got the foundations right
Perhaps this is all a bit tedious, knowing it’s going to go wrong and having to have a plan to bring it back inline. But in 43 years of coding, I’ve worked with many real developers who equally need just as much hand-holding! So nothing new
Just go with the vibe and accept that’s how it is - until Replit improve things.
As for buckets, I am trying to go the full Supabase route for db, auth and file storage. I’ve got the first two bits done, but not tried buckets yet.
I successfully implemented it for my Jewelry Designer for IZIOS, its not live and its far from done. Add to your replit rules that you don’t want it interacting with AWS at all in the future and only use the object storage. I copied the object storage info into my chat, had it code to that information and then It started saving all the generated images to the object storage bucket.
I haven’t used the jewelry designer in about 12 days, I’m building out a local jewelry builder I trained myself using SDXL that will generate the images and send them over to object storage through the api I built, I have a Mac mini cluster that I will have so people can generate free jewelry designs using the diamonds we sell, it was impossible to do a cost effective route with Dalle3 its going to bury me if people sit there all day generating images lol.
I do get why Replit are plugging in lots of tools - user auth, neon DB,… now storage buckets, and soon domain name purchasing. This is all easy revenue for them.
But do you wonder if, for power users, sticking with these is the right strategy?
Or should we free ourselves and find best of breed tools?
Personally for me, Supabase ticks all my DB/auth/bucket boxes.
I think the goal will be to build most of the app on here for launch, get it working in real world and then porting to the services should not be the end of the world anyway, at that point hire someone to help make the transition to a more robust setup. I think you can handle thousands of clients with what replit give you, the issue would be when scaling to many more than that, that might be a problem.
I’ve added Replit Object Storage few times to my apps, usually it did it quite smoothly. I told agent to use it and pasted the code example (typescript) from Replit docs, then it used right packages and got images saved to storage.
I also added the sharp library to resize the served images from storage to smaller size.
I think its nice that Replit has many tools available out of the box, so I dont need to configure them from multiple sites ( etc. Neon, Google Object storage.)
I tried for hours and hours to get Replit Object Storage to work. It never did.
I switched to imagekit, and it worked right away, no issues whatsoever.
The replit issue is that there is not enough documentation in the SDK for it to work. I don’t think replit knows how to get it to work right, since there is limited documentation on it.
Thats been my thoughts, I will say now that it is working in my application, I have no problem changing, adding features to it, or creating new folders via the agent. Just that initial hurdle is always a slog.
