Not today. The current version is purely procedural — it's pattern-matching regex against your prompt, picking the right recipe, plugging in your numbers, and outputting the result. No external API call, no key, no per-use cost, no latency hit.
That means it CAN'T understand free-form prompts like "a viking helmet" or "a coffee table leg" — only the recipes we've coded.
LLM fallback IS on the roadmap. The plan is: try procedural first (instant + free for everything we know), fall back to Claude when the parser misses, with a tight JSON-schema response so the rest of the pipeline doesn't change. No ETA — coming when there's real demand.
That means it CAN'T understand free-form prompts like "a viking helmet" or "a coffee table leg" — only the recipes we've coded.
LLM fallback IS on the roadmap. The plan is: try procedural first (instant + free for everything we know), fall back to Claude when the parser misses, with a tight JSON-schema response so the rest of the pipeline doesn't change. No ETA — coming when there's real demand.