How patentideagenerator works

The autonomous pipeline runs nightly and moves an idea from external signals all the way to a USPTO-ready bundle:

1. Mining— pull the day's arXiv preprints, new USPTO publications, and Hacker News stories matching your domains.
2. Synthesis— Claude reads the corpus through your persona's lens and proposes invention candidates that combine, extend, or sidestep prior art.
3. Triage — every candidate is scored on novelty, patentability, and viability; only those above your approval threshold reach /today.
4. Drafting — once you approve, the engine writes claims, spec, and abstract using the @innovation/* patent packages.
5. Bundle — outputs are validated (antecedent basis, claim numbering, fee math) and packaged as a single zip.
6. File — you download the bundle and submit it in USPTO Patent Center yourself (we never submit on your behalf), then record the application number on your /filings page.

Source-of-truth doc: AUTONOMOUS-PIPELINE.md in the project repository.

patentideagenerator is software, not a law firm. We do not give legal advice — and we never file on your behalf. The platform prepares a USPTO-ready bundle; you submit it yourself in Patent Center as the pro-se inventor (or hand it to your patent attorney to file). You are the named applicant, and you retain control of every approval. This posture is grounded in Sperry v. Florida, 373 U.S. 379 (1963), which holds that federal patent practice preempts state UPL rules for registered patent practitioners and, by extension, for inventor-driven self-representation supported by software tools.

Source-of-truth doc: UPL-ANALYSIS.md in the project repository.