PatentNode reads the Office Action, fetches every cited reference, scores every rejection, and drafts amendment and argument responses. The judgment calls stay with you.
Every stage runs on a model purpose-built for US Office Action responses.
The structured analysis waits for your review before any draft is generated. Edit the reference mechanism, the claim mapping, the combination motivation, the amendment risk; the drafting models fire on your validated brief, not the AI's first guess.
Five specialized models, each tuned for one stage of the practitioner's work: rejection assessment, claim-and-risk analysis, amendments, arguments, and firm-precedent retrieval. Long-trained on the public USPTO corpus, working as an integrated matrix, not a general-purpose API wrapper.
Each rejection becomes a structured bundle: reference mechanism, claim-element mapping, combination motivation, amendment risk. Built for skim-and-correct, not paragraph-deep prose.
The polished analysis report renders in English and Traditional Chinese, with statute citations matched to the right jurisdiction. Built for cross-border prosecution practices — not a machine-translation afterthought.
Pick the technical area (mechanical, electrical, software, biotech, chemical, business-method) and the register (formal-traditional, concise, structured, register-mirroring); the argument and amendment models steer accordingly. Same dispositive analysis, your voice on top.
The agent does the legwork; you make the judgment calls.
Every rejection extracted with statute (§101 / §102 / §103 / §112), examiner reasoning, and cited references.
Patent Center, Google Patents, NPL DOI lookup; cited references are pulled and parsed automatically.
Per rejection: reference mechanism, claim mapping, combination motivation, amendment risk. The bundle waits for you.
Review and edit the analysis in the workbench. The drafting models only fire after you confirm.
Amendment and argument language drafted from your validated brief; rendered as a polished EN + zh-TW report for the client.
Below: one rejection's full structured analysis, rendered side-by-side in English and Traditional Chinese.
Smith teaches a temperature-controlled reservoir; Lee teaches a pressure-controlled valve. The examiner reads Lee's valve as supplying Smith's reservoir-level signal.
Claim 1 element (c) — 'a feedback loop responsive to fluid pressure' — is mapped to Lee Fig. 3 element 28 in the rejection.
The examiner cites MPEP §2143 (A) — combining known elements for predictable result. The strongest counter is that Smith and Lee target different problem domains.
Narrowing claim 1 to specify 'closed-loop pressure feedback below 5 psi' avoids Lee's broader teaching without surrendering the original commercial scope.
Sample-mode only — no upload, no private content, no PII. See the full pipeline on a representative Office Action from the public USPTO corpus.