Ask patent questions in plain language. Get measured answers.
A Claude Code skill + zero-dependency CLI for patent research: counts, leaderboards, trends, portfolios, and full landscape reports over worldwide patent publications — no API key, no pip install (Python 3.8+ stdlib only).
Three things this tool measured on day one (2026-08-18):
- The top filer of
"agentic AI"patents isn't an AI lab — it's Salesforce, and Citibank is #4. - OpenAI has 85 patent publications. Anthropic has 19. NVIDIA has 33,226.
- Filings mentioning
"large language model": 182 (2022) → 64,201 (2025) — a ~350× ramp in three years.
Full analysis: reports/2026-08-18-agentic-ai.md
/plugin marketplace add kevin9327/patent-intel
/plugin install patent-intel@patent-intel
Then just ask Claude things like:
- "Who is patenting retrieval-augmented generation?"
- "Has anyone patented on-device KV cache compression?"
- "Compare OpenAI, Microsoft and Salesforce on AI agent patents"
- "Give me a patent landscape report on humanoid robotics"
git clone https://github.com/kevin9327/patent-intel
cd patent-intel
python plugins/patent-intel/skills/patent-intel/scripts/patent_search.py leaderboard '"agentic AI"'Copy plugins/patent-intel/skills/patent-intel/ into ~/.claude/skills/.
| Command | Question it answers |
|---|---|
count '"solid state battery"' |
How many patent publications mention this? |
search '"KV cache"' --sort new --num 10 |
What was filed recently? (with links) |
leaderboard '"retrieval augmented generation"' |
Who files the most in this space? |
trend '"large language model"' --from 2019 |
Is patenting here accelerating? |
portfolio "Anthropic" |
What does this company patent? |
compare '"AI agent"' --assignees "OpenAI,Microsoft,Salesforce" |
Who leads among these? |
report --domain agentic-ai --out report.md |
Full landscape report for an area |
domains |
List curated domain packs |
selftest [--live] |
Verify the tool works |
Filters on most commands: --country US|KR|EP|WO|..., --status GRANT,
--after 2024, --assignee, --inventor, --json for machine-readable
output.
Curated query sets for areas people actually ask about — one command gets you a full landscape report (volumes, top filers, trend, fresh filings):
agentic-ai · llm-core · rag · ai-inference · ai-chips ·
humanoid-robotics · autonomous-driving · ev-battery ·
industrial-inspection · digital-health · quantum-computing · space-tech
See domains.md for how packs are designed and how to add one.
data/snapshots/ and reports/ grow over time
(weekly workflow + manual runs): dated leaderboards, trends, and lab watchlist
counts. Over months this becomes a diffable public record of who is patenting
what in AI — something a one-off search can't give you.
The zero-key backend is Google Patents' public search endpoint, so the CLI
behaves like a considerate guest: 24h response cache, 2s minimum spacing
(PATENT_INTEL_DELAY to raise), and honest back-off messages when Google
rate-limits a bursty IP (it does; wait 10-30 minutes). Interactive research is
fine — bulk collection is not what this backend is for. For bulk or production
use, official free-key APIs (USPTO, EPO OPS, KIPRIS, BigQuery) are the right
tool: see
sources.md.
Patent publications are public documents; this tool reads public bibliographic data, stores no personal data, and redistributes only small aggregate snapshots (counts, rankings, links). Numbers are patent publications (not grants or deduplicated families) unless filtered, recent years are undercounted by the ~18-month publication lag, and leaderboard facets are sampled. Nothing here is legal advice — for infringement/freedom-to-operate questions, hire a patent attorney.
- v0.2: official API backends behind env keys (
USPTO_API_KEY,EPO_KEY/EPO_SECRET,KIPRIS_API_KEY) for bulk-safe collection - Chart images in reports; more domain packs; family-level dedup
한국 특허청(KIPRIS Plus) 백엔드가 로드맵에 있습니다. 지금도
--country KR로 한국 공보를 검색할 수 있습니다:
python plugins/patent-intel/skills/patent-intel/scripts/patent_search.py search '"결함 검출"' --country KR --sort newMIT