See attached PDF for project proposal. Following message will have additional pdf going into technical specs and blueprint.
#Custom Diagnostic/IT/Repair USB Platform
eternal ocean
pseudo swanBOT
Hi Zartanian! Thanks for sharing your project proposal. However, this appears to be a project planning/discussion post rather than a specific coding question, which is what this forum is designed for.
To get the best help here:
- Do you have a specific programming problem or technical issue you're stuck on? If so, please share the relevant code and error details directly in your post (not as attachments).
- If you're looking for general feedback on your project proposal, this might be better suited for a project showcase or general discussion forum.
Feel free to come back if you hit a specific coding challenge! 👍
eternal ocean
Blueprint as promised. @lucid bison
lucid bison
The smarter dev bot is actually capable of pulling all the parts together too but won't have the same "wow" factor as a local agent manager that has full access to a VM.
Due to this nice "gold rush" period of companies courting us nerds for attention, it seems like a crime to pay anything until you have a good result? 🤷♂️
Another perk of a local agent manager is you want to be able to work on this and expand it right?
Well at each step the agent manager is trying to make this a code learning experience (that's somewhat baked into it) so it will even make scripts (calling them "tools") to tackle things locally.
Just roll with that. Encourage it to make lots of scripts/tools both of you can tweak/run as needed.
If you never pay and stay in that self-service mindset you'll be in less of a panic if the AI is expensive/all tied up/offline later?
eternal ocean
@lucid bison Okay, sorry for vanishing, had personal stuff come up, but I could use your help since you seem more familiar with this stuff. My current environment is VS Code with Continue.dev, qwen3-coder:30B local and qwen3-coder:480B-cloud. I cannot for the life of me get Continue.dev to be capable of actually creating files in the workspace. It can see it, it believes it's creating them, shows no errors even in debug logs, yet nothing actually occurs.
every bit of code it creates has either this create file option, or an apply option, however create file does nothing at all, and apply only adds the code to whatever file is open manually. so when it doesn't tell me what file to make, or where to put it, there are issues.
lucid bison
If the Google Antigravity public preview I'm using is off the table, let's ask @pseudo swan what alternatives you can explore to have a local AI agent that can run in an "Agent Manager" mode creating tool scripts and running unattended on a VM that it has sudo access on. There is a 60% chance I know the name of at least one suggestion but you'll be on your own here in terms of troubleshooting.
eternal ocean
after asking around, with both people and ai, my current setup seems to be the best I can do without paying money. I just can't get it to actually function correctly like it's supposed to. qwen3-coder:30B is among the best free options for local-hosted coding that I can actually run on my machine. qwen3-coder:480b-cloud is also amazing. the issue is, they are both writing code but being completely unable to implement it.
lucid bison
Well the Google Antigravity public beta preview dropped in early Feb?
So not a lot of "experts" have even tried it much less tested the Agent Manager option which is where that "wow" factor is.
However, if you don't want Google, there's an open source alternative (two) called OpenHands that will mimic the same thing, it'd just install on the remote VM as a docker container and you'd leave it to sort tasks out after giving it API access to some free AI agents or local access to an agent.
Open Interpreter is pretty close to the same idea, it's got agent API access too, but takes a lot more setting up and isn't as nice (CLI interface).
Don't get me wrong, if you have VSCode you'll panic that you're wasting disk space on a clone.
Antigravity even imports VSCode data. So the panic isn't crazy.
But I assure you the free work you can do with the Gemini models in the Agent Manager mode will make the disk space A-OK.
In fact, step one requires some VSCode experience. CTRL+SHIFT+P (or C) is needed to get the command pallet up to SSH connect to the VM.
I do a lot of web stuff myself, so I can't picture not letting the AI Agent take over my web browser to have it test code before it calls the work done. 🤓
Oh and if you're an old person like me who is UX blind, the "Open Agent Manger" text at the top of the window is where you go for the "cruise" mode.
