OA Agent

An AI agent that reads the cited prior art, maps each claim limitation to the examiner's evidence, and recommends argue or amend — rejection by rejection, with a citation you can check for every conclusion.

Self-host it Runs on your machine — there's no hosted instance.
Solve This OA For Me
Session #1847

I received a §103 rejection on claims 1-12. The examiner is combining Smith (US 10,234,567) and Jones (US 2021/0045678). Here's the Office Action and my application.

OA_2024-0892.pdf
App_16-789012.pdf
OA_2024-0892.pdf 847 KB read
App_16-789012.pdf 1.2 MB read
Used tool: Grep

I've analyzed the Office Action. Here's my assessment of each rejection:

§103 Rejection — Claims 1-5 (Smith + Jones)

Smith discloses a "multi-layer authentication module" (col. 5, ll. 12-34) but does not teach the claimed dynamic threshold adjustment. The examiner relies on Jones ¶[0067] for this element, but Jones describes a static threshold in a different context (batch processing, not real-time).

Recommendation: ARGUE — Confidence: HIGH

§103 Rejection — Claims 6-12 (Smith + Jones)

The combination of Smith and Jones does fairly teach the "data aggregation pipeline" of claim 6. However, amending to incorporate the "priority queue with weighted scoring" from dependent claim 9 would distinguish over both references. Spec support at ¶[0042]-[0044].

Recommendation: AMEND — Confidence: MEDIUM

See how it works in 40 seconds

Capabilities

The reading and mapping, done for you

Every Office Action gets a thorough first pass. The AI reads the prior art, maps the claim elements, and flags what matters so you can focus on strategy.

Every rejection, mapped

Parses the full Office Action and classifies every rejection by type, mapping each one to the specific claims and prior art references cited.

Prior art, actually read

Reads cited references and follows the examiner's specific citations to the relevant paragraphs and columns. Cross-checks whether the cited passages actually teach what the examiner claims they teach.

Claim elements, mapped to references

For every rejection, maps each claim limitation to the specific passage the examiner relies on. When multiple references are combined, shows exactly which element comes from which reference.

Antecedent basis, checked

Scans every claim for missing antecedent basis, dangling references, and indefiniteness issues. Flags problems before the examiner does.

Argue, amend, or both

Looks for arguments first, only suggesting amendments when arguing alone isn't enough. You get draft argument language and, where needed, suggested claim amendments with spec support.

Check every conclusion yourself

Every step is visible in real time. See which document is being read, which passage is being analyzed. Click any cited source to jump straight to it and verify the conclusion yourself.

Jurisdictions

Works with patent offices worldwide

The AI understands rejection types across all major patent jurisdictions. Upload an Office Action from any of these offices and get a strategy that speaks the right language.

United States (USPTO)

  • §101 Subject Matter Eligibility
  • §102 Anticipation (Novelty)
  • §103 Obviousness
  • §112(a) Written Description / Enablement
  • §112(b) Definiteness
  • and more

European Patent Office (EPO)

  • Art. 52 Patentable Inventions
  • Art. 54 Novelty
  • Art. 56 Inventive Step
  • Art. 83 Sufficiency of Disclosure
  • Art. 84 Clarity & Support
  • Art. 123(2) Added Subject Matter
  • and more

China (CNIPA)

  • Art. 22.2 Novelty
  • Art. 22.3 Inventiveness (Obviousness)
  • Art. 25 Non-Patentable Subject Matter
  • Art. 26.3 Sufficiency of Disclosure
  • Art. 26.4 Claim Clarity & Support
  • Art. 33 Amendment Beyond Original Scope
  • and more

Japan (JPO)

  • Art. 29(1) Novelty
  • Art. 29(2) Inventive Step
  • Art. 36(4) Enablement
  • Art. 36(6) Claim Clarity & Support
  • Art. 29bis Enlarged Novelty
  • and more

South Korea (KIPO)

  • Art. 29(1) Novelty
  • Art. 29(2) Inventive Step
  • Art. 42(3) Enablement
  • Art. 42(4) Claim Clarity & Support
  • and more

PCT (International Phase)

  • Art. 33(2) Novelty (Chapter II)
  • Art. 33(3) Inventive Step (Chapter II)
  • Art. 5 PCT Written Opinion Objections
  • Rule 66 ISA Written Opinion Responses
  • and more

India (IPO)

  • Sec. 3 Non-Patentable Subject Matter
  • Sec. 2(1)(j) Novelty (Prior Art)
  • Sec. 2(1)(ja) Inventive Step
  • Sec. 10(4) Sufficiency of Disclosure
  • and more

Canada (CIPO)

  • Sec. 2 Subject Matter / Utility
  • Sec. 28.2 Novelty (Anticipation)
  • Sec. 28.3 Obviousness
  • Sec. 27(3) Sufficiency of Disclosure
  • and more

Don't see your jurisdiction? The AI has broad knowledge of patent law across jurisdictions. Upload your Office Action and it will adapt.

FAQ

OA Agent FAQ

Does the AI hallucinate? How can I trust the analysis?

The OA Agent works directly from your uploaded documents, not from memory or training data. Every conclusion cites a specific passage, and those citations are verified against the source. You can click any reference to jump to the passage and check it yourself. Every claim is traceable and verifiable.

What types of rejections does it handle?

All major rejection types across all jurisdictions. Novelty, obviousness, subject matter eligibility, written description, definiteness, and their equivalents at the EPO, CNIPA, JPO, KIPO, and PCT. Upload your Office Action and the AI adapts to the relevant framework.

Can it draft the actual response?

Think of it as a thinking tool, not a drafting tool. You get a complete analysis with recommendations, draft arguments, and suggested amendments. It's designed to help you work through the response, not to produce a filing-ready document. You make the final calls and write in your own voice.

How does it compare to other AI patent tools?

Most AI patent tools focus on drafting, proofreading, or prior art search. This is specifically for Office Action responses. You upload an OA and your documents, and the AI reads the cited references, maps claim elements, and produces argue/amend recommendations. We're not aware of another tool that does this end to end.

Why not just use ChatGPT or Gemini?

You can paste an Office Action into ChatGPT, but it can't read your actual prior art PDFs, verify its own citations, or maintain a structured workspace of your documents. This is a specialized AI agent with tools to read, search, and cross-reference your actual files. It follows the examiner's specific citations to the relevant passages and checks them. A general-purpose chatbot gives you a plausible-sounding answer. This gives you a verifiable one.

Self-host it

Next: check your amendments

Once you've decided what to amend, the Antecedent Basis checker reads the amended claim set and flags the §112 problems that get introduced when claims are edited under time pressure.

About Antecedent Basis