@DeepSeekHarness Great write-up. Two things that held up for us: (1) progress checks belong in code, not in a model's judgment. "Same red result twice = stop and ask" saved hours of blind retries. (2) The biggest delay wasn't compute but the answer path back from the human.
@jacekviews@themishra4402 this is when you dint know they have published their models under MIT license you can use for example a host in your coubtry running that model
@themishra4402 deepseek even killed in benchmarks but it is very fast its conclusions are made too fast you have to steer it precisely rathen than with opus just my experience but i use it for 2nd opinion and builidng
@PerceptualPeak i just say trust is good controll is better. this should have been preventable. but me also have daily backups and even offsite backups :-)
@TonyXavier_ hmm well cuz my coding agents work in my loop fabric on my server who runs them on desktop? u need an adapter to add models and then you let the fabric build
@hazyhaarlab@sattyyouneed this is how you did it 6 months ago now you need one orchestrstor discussing your reqs rest you let habdle your loop engineering fabric
@lightsilver323 https://t.co/qmsyAJIWcn next step is to move away from saas to self hosted website on node.js dont need wirdpress and thousand plugins anymore