Quick question before I weigh in properly—are you looking to save money by running something locally, or are you fine with a monthly sub? If you're on a budget, DeepSeek Coder V2 or Mistral AI Codestral 22B are honestly solid for complex logic without the high price tag. Just curious what kind of monthly spend you're aiming for tho?

.

@Reply #3 - good point! lol. Late to the party but I've been obsessed with getting my async logic perfect lately and it's been such a wild ride. I spent like three days straight trying to debug a race condition in a custom decorator for my backend project and it was a total nightmare because the AI kept hallucinating library internals. I finally found a workflow that actually respects the logic, but it took a lot of trial and error! It's so satisfying when you finally get that clean, non-blocking code running without it breaking your auth flow. Reliability is literally everything when you're dealing with multi-layered decorators. I'm curious tho, are you looking for something that can index your entire repo at once, or is your code modular enough that you only need context for a few files at a time? Also, are you more focused on raw logic precision or do you need it to follow a very specific style guide for your production environment?

I've found Anthropic Claude 3.5 Sonnet much more reliable than OpenAI GPT-4o for complex decorators. Claude is way more precise with logic, and Cursor AI Code Editor handles the context best.