I think being able to step back from a bad decision is mutually exclusive with being a CEO of a huge company
2 Likes
Best places to watch this wreck play out, and discover alternative inference providers/approaches?
- Meta: Usage based billing model Ā· Issue #319403 Ā· microsoft/vscode Ā· GitHub
- https://www.reddit.com/r/GithubCopilot/
- Issues Ā· microsoft/vscode Ā· GitHub
Was hoping to wrap-up and share https://control.makergalaxy.com, but thatās on hold now.
This is why Iāve been spending more and more time on my āin houseā AI stack. It will never be a full Claude replacement, but Iāve found I can do quite a bit with it.
I donāt have a lot of money to throw at it, so itās just a single linux server with a single 5060ti 16G card on it (for now). Iām using opencode (which is where Iāve been spending a lot of time making changes). The idea is to allow opencode to use claude api credits for initial planning and āharderā tasks, then push the smaller grunt agents to my local AI. Right now, Iām even having all the harder tasks ran on it using one of the newer MoE models. Itās working OK. Itās just slower. Which is fine for my personal work.
āwork workā is paying for claude enterprise. I have a personal claude plan Iāll use when the opencode gets stuck on an issue or is taking too long. Iāve only hit my limits a few times on my $20 plan using this setup.
2 Likes