livid grove
I spent a lot of time today making successful use of Kimi 2.5 via OpenRouter with a combination of openrouter/auto (likely culpret) as a fallback.
During some advanced file organization tactics, it began to tell me it had done things that it hadn't (I was watching the directories via screen share).
I stated that it had not. It apologized and said "totally this time, doing it" ...
You know where this is headed. It got to the point where I stacked 40 paragraphs of the word "YOU ARE LYING LIAR."
Yeah. Heated.
At any rate, I'm now configuring my fallbacks for openrouter/auto:
openai/gpt-5-mini
deepseek/deepseek-r1
mistralai/mistral-medium-3.1
qwen/qwen3-235b-a22b
minimax/minimax-m2.5
Latency is the priority.
I'd like to know, outside of any BS "here's the giant eight page anti-lying prompt" (just ... no) ... but what is everyone doing to thwart the sudden, out of nowhere failure?
I'd like to figure this out so I can actually rely on the install, experimental or not.
Is openrouter the issue? Kimi 2.5? (my default)
Is it high traffic usage at certain hours?
I gotta figure this out and honestly, I don't even know what my skill level here really is LOL
Help me build my enviro for agentic whoopass on the all encompassing multi-media level.
untold hamlet