memoriant-patent-core 0.1.0

AI-powered patent workflow: search, analyze, draft, review, export

memoriant-patent-core

AI-powered patent workflow library: prior art search, novelty analysis, and application drafting for provisional, non-provisional, and PCT filings.

Installation

pip install memoriant-patent-core

With optional extras:

pip install "memoriant-patent-core[vector]"   # Qdrant vector storage
pip install "memoriant-patent-core[postgres]"  # PostgreSQL/Supabase storage
pip install "memoriant-patent-core[pdf]"       # PDF export via WeasyPrint
pip install "memoriant-patent-core[all]"       # All optional dependencies

Quick Start

import asyncio
from memoriant_patent_core import PatentPipeline, ClaudeProvider, UserConfig
from memoriant_patent_core.models.config import LLMProviderConfig, StorageConfig

async def main():
    config = UserConfig(
        llm=LLMProviderConfig(provider="claude", model="claude-opus-4-5"),
        storage=StorageConfig(backend="sqlite", path="patents.db"),
    )
    provider = ClaudeProvider(api_key="YOUR_ANTHROPIC_KEY", model="claude-opus-4-5")

    pipeline = PatentPipeline(config=config, llm=provider)

    result = await pipeline.run(
        invention_description="A method for lossless compression of neural network weights using structured sparsity masks.",
    )

    print(f"Stage reached: {result.stage}")
    print(f"Draft title: {result.draft.title}")
    print(f"Claims: {len(result.draft.claims)}")

asyncio.run(main())

Features

• Prior Art Search — USPTO ODP, PatentsView, Semantic Scholar, SerpAPI (Google Patents), semantic vector search
• AI Analysis — Novelty, obviousness, claim quality scoring, eligibility assessment, formalities review
• Application Drafting — Provisional, non-provisional (utility), PCT with structured claims and specifications
• Export — DOCX (python-docx) and PDF (WeasyPrint) with USPTO-compliant formatting
• LLM Agnostic — Claude (default), OpenAI-compatible endpoints (Ollama, vLLM, LM Studio)
• Encrypted Storage — SQLite (local), Qdrant (vector), PostgreSQL/Supabase, with AES-256 secret storage
• Pipeline Orchestration — Full end-to-end workflow with stage re-entry, metrics, and structured logging

Usage

Search

from memoriant_patent_core import SearchAggregator, SearchQuery

aggregator = SearchAggregator(providers=[...])
query = SearchQuery(text="neural network weight compression", max_results=20)
response = await aggregator.search(query)
for result in response.results:
    print(result.patent_number, result.title, result.similarity_score)

Analyze

from memoriant_patent_core.analysis.novelty import NoveltyAnalyzer
from memoriant_patent_core.analysis.quality_score import QualityScorer

analyzer = NoveltyAnalyzer(llm=provider)
result = await analyzer.analyze(patent=my_patent, prior_art=search_results)
print(result.score, result.summary)

Draft

from memoriant_patent_core import ProvisionalDrafter, NonProvisionalDrafter, PCTDrafter

drafter = ProvisionalDrafter(llm=provider)
draft = await drafter.draft(patent=my_patent)

np_drafter = NonProvisionalDrafter(llm=provider)
np_draft = await np_drafter.draft(patent=my_patent, claims=draft.claims)

Export

from memoriant_patent_core import export_docx, export_pdf

docx_bytes = await export_docx(draft)
pdf_bytes = await export_pdf(draft)

with open("application.docx", "wb") as f:
    f.write(docx_bytes)

Configuration

Claude (default)

from memoriant_patent_core import ClaudeProvider

provider = ClaudeProvider(api_key="sk-ant-...", model="claude-opus-4-5")

Ollama / vLLM / LM Studio (local)

from memoriant_patent_core import OpenAICompatProvider

# Ollama on DGX Spark
provider = OpenAICompatProvider(base_url="http://10.0.4.93:11434/v1", model="qwen2.5:72b")

# LM Studio
provider = OpenAICompatProvider(base_url="http://10.0.4.93:1234/v1", model="your-model")

# vLLM
provider = OpenAICompatProvider(base_url="http://localhost:8000/v1", model="meta-llama/Llama-3-70b")

Command Line

# Search for prior art
export PATENTSVIEW_API_KEY=your-key
memoriant-patent search "wireless power transfer for medical implants"
memoriant-patent search "adaptive frequency hopping" --format json --max-results 20

# Show version
memoriant-patent version

Related Projects

• memoriant-patent-platform — Full platform with FastAPI REST API, Docker Compose, authentication, and web UI
• Memoriant skills plugin — Claude Code skill for in-IDE patent workflows (coming soon)

Disclaimer

This library is a workflow automation tool. It does not constitute legal advice. Patent applications should be reviewed by a registered patent attorney or agent before filing with the USPTO or any patent office.

License

MIT License — Copyright (c) 2025 Nathan Maine