When i disconnect the network and activate ollama and bolt.diy
The files and code are not being generated, although the request is sent to the model
1 Like
Am i the only one that encounter it?
For some reason it needs network online to work,
When connected it works perfectly, but when disconnecting network - no files, no virtual system
Likely due to the model you are running. Only “Instruct” models 7B+ generally work with artifacts (terminal, files, etc.).
And if you mean completely offline, then it just won’t work. The current state of Bolt.diy can’t be 100% cached locally and the Web Container API calls to the web.
Thanks for the reply,
I was trying lamma 70B Instruct as well,
I am talking about 100% offline,
Why cant it be 100% offline (for the package managers?) or for other reasons ?
Part of it is that full PWA support needs to be added with caching (easy), but also the Web Container (closed source) calls to its API. That component can’t be run offline, so it would need to be replaced (different terminal and backend). Not so easy.
1 Like
so it means that bolt.diy is not really private, it uploads the code to bolt.new servers for the backend to do it’s magic, and ot also means that bolt.diy cannot work without the company server.
right ?
Not quite right. There are integrations in the backend for Python (just one example), terminal features, etc. Look for Python in the code and you won’t find it, but you type in python in the terminal and it just works.
Stackblitz created Bolt.new and open sourced it, but the Web Container is their proprietary product they make money off of. And they are 100% entitled to make money on the products they make.
The calls back to the API make the terminal function without giving away “how” it functions. I think it’s just a clever way to obfuscate the code. And there’s probably a lot more to it that I don’t know about or understand.
That’s fine and all, none of us know it all, but I would be careful about making pointed false claims. It’s not really “uploading” data or using it in the way you imply.
its obviously more then ok to make money on the company invention (but if they do not make money off the opensource code - then there is no profit considerations here).
but it is still sending data to their server, i can definitely find out what is being sent (this is my line of work), i will test it and reply back, but i see no way for a technology to work as a server side on a code without having access to the code !
It’s an API call to receive data more than send data.
I’ve voiced my opinion to replace it entirely, but that will be some work. I do have a side project to replace the Web Container/Terminal, and I would consider making it an opt-in feature in settings. But I have to finish my development, testing, and integration. This has been mentioned here.
has there been any progress on this ? anyone have direction on how to make it work when completely offline ? i would really want to use it
Hi @shay.games, no, there is no process regarding this and most likely also will not happen. See also: Replacing webcontainer.io with gptengineer - #3 by ColeMedin
currently i am using Roo-Cline & Aider which are pretty good replacements, but i thought bolt would be a nice addition to the development/architecture/design tools,
i would invest developing the server side if i had to, but since you say only web containers are the server side for bolt,diy and it can never be hosted locally - then i will give-up on this path
thanks and good luck