学术文献智能处理工具集。覆盖文献从元数据到结构化知识的全流程，帮助科研人员自动完成文献检索、PDF 解析、知识提取、证据检索、Discussion 生成和综述撰写。
Zotero → Metadata → PDF → Full Text → Analysis → KB → Evidence → Discussion → Review
(m1) (m2) (m3) (m4/5) (m6) (m7) (m8)
| 模块 | 名称 | 功能 |
|---|---|---|
| m1 | litmind-zotero | Zotero 连接器 — 导出文献元数据 |
| m2 | litmind-parser | PDF 解析 — 清洗噪声、识别章节结构 |
| m3 | litmind-analyzer | LLM 知识提取 — 研究问题/方法/变量/发现/声明 |
| m4 | litmind-knowledge | 知识库 — SQLite + ChromaDB 存储与检索 |
| m5 | litmind-chat | 科研问答 — 基于知识库的自然语言问答 |
| m6 | litmind-evidence | 证据检索 — 支持/反对/中性证据分类 |
| m7 | litmind-discussion | Discussion 生成 — 7 节结构化草稿 |
| m8 | litmind-review | 综述生成 — 主题聚类/共识/争议/空白+全文 |
pip install litmind
# 按需安装依赖
pip install litmind[pdf] # PDF 解析 (pymupdf)
pip install litmind[llm] # LLM 分析 (anthropic + openai)
pip install litmind[kb] # 知识库 (sqlalchemy + chromadb)
pip install litmind[all] # 全部依赖export ANTHROPIC_API_KEY=sk-... # Claude provider (m3, m5, m6, m7, m8)
export OPENAI_API_KEY=sk-... # OpenAI provider (可选)从 Zotero 本地 SQLite 数据库读取期刊论文元数据。
python scripts/cli.py export -o papers.json
python scripts/cli.py statsfrom litmind_zotero import discover_database, export_all
papers = export_all(discover_database())输出： PaperMetadata — key, title, authors, year, doi, journal, pdfPath, tags, collections
从 PDF 提取全文 → 自动清洗（页眉/页脚/页码/重复）→ 识别标准章节。
python scripts/parse.py single paper.pdf -o parsed.json
python scripts/parse.py batch --from-zotero papers.json -o parsed/from litmind_parser import parse_pdf
result = parse_pdf("paper.pdf")
print(result.sections.abstract[:200])输出： PaperContent — fullText + sections (abstract/intro/methods/results/discussion/conclusion)
将论文全文通过 LLM 提取为结构化科研知识。
litmind-analyze parsed.json -o analysis.json
litmind-analyze paper.json --provider openai --model gpt-4ofrom litmind_analyzer import analyze_paper
from litmind_analyzer.providers import AnthropicProvider
provider = AnthropicProvider()
result = analyze_paper(paper_content, provider)
print(f"研究问题: {result.researchQuestion}")
print(f"方法: {result.methods}")输出： PaperAnalysis — researchQuestion, studyDesign, participants, methods, statistics, variables, claims, limitations, keywords
基于 SQLite + ChromaDB 的本地科研知识库，存储论文分析结果，支持语义检索和结构化查询。
litmind-knowledge add analysis.json
litmind-knowledge search "flatfoot kinematics"
litmind-knowledge statsfrom litmind_knowledge.service import KnowledgeBase
kb = KnowledgeBase()
kb.add_paper(analysis_dict)
results = kb.semantic_search("MTP ROM flatfoot", top_k=10)面向知识库的自然语言问答。自动分类问题类型、检索相关证据、生成带引用的回答。
litmind-chat "What studies support the link between flatfoot and MTP ROM?"
litmind-chat "Show me papers using SPM analysis"from litmind_chat.service import ResearchChatService
chat = ResearchChatService(kb=kb, llm_provider=provider)
response = chat.ask("Does flatfoot increase forefoot motion?")
print(f"Answer: {response.answer}")
print(f"Sources: {len(response.supportingPapers)} papers")输入一个科学观点，自动检索知识库中的支持证据、反对证据和中性证据，评估证据强度。
python scripts/evidence.py "Flatfoot increases MTP ROM"
python scripts/evidence.py "Carbon plate shoes improve jump performance" --jsonfrom litmind_evidence import EvidenceFinderService
ev_service = EvidenceFinderService(kb=kb, llm_provider=provider)
result = ev_service.find_evidence("SPM is more sensitive than peak-value analysis")
print(f"Strength: {result.evidenceStrength}") # Strongly / Moderately / Weakly Supported
print(f"Support: {len(result.support)} papers")
print(f"Oppose: {len(result.oppose)} papers")证据强度分级： Strongly Supported | Moderately Supported | Weakly Supported | Mixed Evidence | Insufficient Evidence
输入研究结果，自动检索相关文献，生成 7 节结构化 Discussion 草稿。
python scripts/discussion.py \
--topic "Footwear stiffness effects on biomechanics" \
--results "High stiffness shoes increased MTP ROM" \
--results "No significant difference in ankle sagittal ROM"from litmind_discussion import DiscussionGeneratorService, DiscussionInput
service = DiscussionGeneratorService(evidence_service=ev_service, llm_provider=provider)
inp = DiscussionInput(studyTopic="Footwear stiffness", results=["...", "..."])
result = service.generate_discussion(inp)
print(result.discussionDraft)7 个 Section：
- Main Finding Interpretation
- Supporting Evidence
- Contradictory Evidence
- Potential Mechanisms
- Practical Implications
- Study Limitations
- Future Directions
输入一个研究主题，自动完成文献发现、主题聚类、趋势分析、共识与争议识别、研究空白发现，生成结构化综述全文。
python scripts/review.py "Flatfoot Biomechanics"
python scripts/review.py "SPM in Biomechanics" --jsonfrom litmind_review import ReviewGeneratorService, ReviewInput
service = ReviewGeneratorService(kb=kb, evidence_service=ev_service, llm_provider=provider)
inp = ReviewInput(topic="Flatfoot Biomechanics", max_papers=50)
result = service.generate_review(inp)
print(f"Themes: {len(result.researchThemes)}")
print(f"Consensus: {len(result.researchConsensus)}")
print(f"Controversies: {len(result.researchControversies)}")
print(f"Gaps: {len(result.researchGaps)}")
print(result.reviewDraft)核心功能：
- ThemeDiscoveryEngine — LLM 自动聚类，提炼 3-7 个研究主题
- TrendAnalyzer — 统计高频变量、统计方法、研究设计、年份分布
- ConsensusAnalyzer — 识别多篇文献一致支持的结论
- ControversyAnalyzer — 发现支持和反对证据并存的争议点
- GapAnalyzer — 找出低频研究方向和空白领域
- ReviewComposer — 逐节生成 8 个 Section 的综述全文
8 个 Section：
- Introduction
- Current Research Landscape
- Major Research Themes
- Evidence Consensus
- Research Controversies
- Research Gaps
- Future Directions
- Conclusion
所有模块的 LLM 生成内容均遵循严格的白名单引用机制：
- LLM 只能引用 Knowledge Base 中的真实 paperId
- 所有引用在后处理阶段校验，未通过的自动丢弃
- 禁止虚构作者、年份、DOI、期刊
在 Claude Code 中可直接调用以下命令：
/litmind-zotero Zotero 连接器
/litmind-parser PDF 解析
/litmind-analyzer 论文知识提取
/litmind-knowledge 知识库
/litmind-chat 科研问答
/litmind-evidence 证据检索
/litmind-discussion Discussion 生成
/litmind-review 综述生成
litmind/
├── src/
│ ├── litmind_zotero/ # m1
│ ├── litmind_parser/ # m2
│ ├── litmind_analyzer/ # m3
│ ├── litmind_knowledge/ # m4
│ ├── litmind_chat/ # m5
│ ├── litmind_evidence/ # m6
│ ├── litmind_discussion/ # m7
│ └── litmind_review/ # m8
├── scripts/
├── tests/
├── .claude/skills/
├── pyproject.toml
└── README.md
MIT