LM Studio powers WoW addon tuning

This is exactly where local models make sense: repetitive, source-heavy coding work where privacy and usage caps matter more than frontier-level reasoning. The catch is that BIS accuracy is a data problem first, so the model should propose candidates while deterministic code decides which source wins.

• –Local coding stacks are getting good enough that quota, not capability, becomes the reason to switch
• –LM Studio's OpenAI-compatible local server makes it easy to plug into editor workflows without rewriting everything
• –Qwen Code Companion plus VS Code is a low-friction path for trying local coding assistants fast
• –For an addon like this, the hardest part is source reconciliation, ranking rules, and validation, not generating text
• –If the local model starts hallucinating item priorities, tighter prompts and lower-temperature sampling will matter more than raw model size