LM Studio users seek RP frontends, models

This is a useful snapshot of where local roleplay has matured: the hardware is no longer the constraint, the UX stack is. The hardest part is finding a frontend that treats memory, lorebooks, and character cards as first-class objects instead of bolted-on features.

• –LM Studio remains the easy on-ramp, but its UX is optimized for model running, not RP workflow.
• –AnythingLLM is closer to a general knowledge workspace than a character-card-first story engine.
• –The post shows demand for a “SillyTavern-lite” layer: less configuration overhead, but still strong memory, card import, and persona handling.
• –On a 4090, the model sweet spot is likely high-quality 30B-class or carefully quantized 70B-class models, with context management mattering more than raw parameter count.
• –The real decision is not just model quality; it’s whether the frontend can keep state coherent over long sessions without fighting the user.