SM Online

What Makes SM Online Different?

SM Online isn't just a management game—it’s a living world powered by the LLM of your choice. While it honors the deep strategic roots of the management genre, it pushes the boundaries of how characters actually "think."

• Brains You Can Swap: You aren't stuck with one AI. Connect to OpenRouter or OpenAI, or run a 100% private server locally via Ollama.
• Persistent Memories: NPCs don’t just follow scripts. They remember your choices, adapt to your management style, and react with context-aware, radiant dialogue.
• Cinematic Depth: Experience a 16-scene narrative prologue that weaves your character's stats and background into the lore of the Kingdom of Mioya.
• Complete Control: From managing a complex weekly schedule to interacting with a lore-rich knowledge base of 1,200+ entries, every decision influences the world around you.

AI & Model Recommendations

We are constantly testing to find the best "brains" for the game. Currently:

• Top Pick: dolphin-mistral-24b-venice-edition (via OpenRouter) has shown excellent reasoning and creative results in our testing.
• Local Pick: llama-3-lumimaid-8b-v0.1 (via Ollama) is a solid local option, though it can occasionally become repetitive.

What are you running? We’d love to hear which models and settings are giving you the best dialogue and behavior. Let us know in the comments!

New in 0.7.0: The Events Update

• Config Console — the settings drawer is gone. Configuration now lives at /config: sidebar + topbar with nine tabs (Overview · Providers · Language Model · Image Model · Audio · Display · Gameplay · Content · Privacy). The topbar gear on every page deep-links to it.
• Vast.ai GPU host — first-class deployment target for Ollama. A "Where it runs" toggle (Local · Vast.ai) lets Vast mode rent and provision a GPU instance, route chat traffic to it, and reap it when idle. Per-mode settings persist independently, so flipping back and forth is non-destructive. ComfyUI on Vast.ai is not in this release — image generation with ComfyUI is local-only for now (point it at your own endpoint URL); cloud ComfyUI deployment is coming in a later release.
• ComfyUI: FLUX support + clearer errors. The Image Model → Workflow picker now offers a built-in FLUX preset alongside SD1.5 and Custom — it loads the diffusion model, text encoders and VAE separately (FLUX checkpoints don't bundle CLIP/VAE), so a FLUX .safetensors works once the t5xxl / clip_l / ae files are staged on the worker. The configured checkpoint now drives either loader, prompt injection follows conditioning nodes like FluxGuidance, and a failed generation surfaces the actual ComfyUI node error instead of waiting out the timeout.

Content: Explicit sexual themes, power dynamics, adult language. All characters are 18+. This is a consensual adult fantasy simulation.