papi

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Paper Reference Assistant (CLI)

论文参考助手（CLI）

Entry point skill. Use

papi

CLI first; MCP RAG tools only when CLI is insufficient.

For specialized workflows, invoke dedicated skills:

```
/papi-ask
```
— RAG queries requiring synthesis
```
/papi-verify
```
— verify code against paper
```
/papi-compare
```
— compare papers for decision
```
/papi-ground
```
— ground responses with citations
```
/papi-curate
```
— create project notes

入口Skill。优先使用

papi

CLI；仅当CLI功能不足时才使用MCP RAG工具。

针对特定工作流，调用专用Skill：

```
/papi-ask
```
—— 需要综合分析的RAG查询
```
/papi-verify
```
—— 验证代码与论文的一致性
```
/papi-compare
```
—— 对比论文以辅助决策
```
/papi-ground
```
—— 基于引用生成可靠回复
```
/papi-curate
```
—— 创建项目笔记

Setup

设置

bash

papi path   # DB location (default ~/.paperpipe/; override via PAPER_DB_PATH)
papi list   # available papers
papi list | grep -i "keyword"  # check if paper exists before searching

bash

papi path   # 数据库位置（默认~/.paperpipe/；可通过PAPER_DB_PATH环境变量覆盖）
papi list   # 查看可用论文
papi list | grep -i "keyword"  # 搜索前先检查论文是否存在

When NOT to Use MCP RAG

何时不使用MCP RAG

Paper name known →
```
papi show <paper> -l summary
```
Exact term search →
```
papi search --rg "term"
```
Checking equations →
```
papi show <paper> -l eq
```
Only use RAG when above methods fail or semantic matching required

已知论文名称 →
```
papi show <paper> -l summary
```
精确术语搜索 →
```
papi search --rg "term"
```
查看公式 →
```
papi show <paper> -l eq
```
仅当上述方法失效或需要语义匹配时才使用RAG

Decision Tree

决策树

Question	Tool
"What does paper X say about Y?"	`papi show X -l summary` , then `papi search --rg "Y"`
"Does my code match the paper?"	`/papi-verify` skill
"Which paper mentions X?"	`papi search --rg "X"` first, then `leann_search()` if no hits
"Compare approaches across papers"	`/papi-compare` skill or `papi ask`
"Need citable quote with page number"	`retrieve_chunks()` (PQA MCP)
"Cross-paper synthesis"	`papi ask "..."`

问题	工具
“论文X关于Y说了什么？”	`papi show X -l summary` ，之后使用 `papi search --rg "Y"`
“我的代码是否与论文匹配？”	`/papi-verify` Skill
“哪篇论文提到了X？”	先使用 `papi search --rg "X"` ，若无结果再使用 `leann_search()`
“对比多篇论文的方法”	`/papi-compare` Skill或 `papi ask`
“需要带页码的可引用引文”	`retrieve_chunks()` （PQA MCP）
“跨论文综合分析”	`papi ask "..."`

Search Commands

搜索命令

bash

papi search --rg "query"              # literal text match (fast, no LLM) — NOT regex by default!
papi search --rg --regex "pattern"    # regex patterns (add --regex explicitly)
papi search "query"                   # ranked BM25
papi search --hybrid "query"          # ranked + exact boost
papi search "query" -p paper1,paper2  # limit search to specific papers
papi ask "question"                   # PaperQA2 RAG
papi ask "question" --backend leann   # LEANN RAG
papi notes {name}                     # open/print implementation notes

bash

papi search --rg "query"              # 字面文本匹配（快速，无需LLM）—— 默认不是正则表达式！
papi search --rg --regex "pattern"    # 正则表达式匹配（需显式添加--regex参数）
papi search "query"                   # BM25排序搜索
papi search --hybrid "query"          # 排序+精确匹配增强
papi search "query" -p paper1,paper2  # 限制搜索特定论文
papi ask "question"                   # PaperQA2 RAG
papi ask "question" --backend leann   # LEANN RAG
papi notes {name}                     # 打开/打印实现笔记

Search Escalation (cheapest first)

搜索升级策略（成本从低到高）

```
papi search --rg "X"
```
— exact text, fast, no LLM

papi search "X"

— ranked BM25 (requires

papi index --backend search

first)

```
papi search --hybrid "X"
```
— ranked + exact boost
```
leann_search()
```
— semantic search, returns file paths for follow-up
```
retrieve_chunks()
```
— formal citations (DOI, page numbers)
```
papi ask "..."
```
— full RAG synthesis

```
papi search --rg "X"
```
—— 精确文本匹配，快速，无需LLM

papi search "X"

—— BM25排序搜索（需先执行

papi index --backend search

）

```
papi search --hybrid "X"
```
—— 排序+精确匹配增强
```
leann_search()
```
—— 语义搜索，返回文件路径以便后续操作
```
retrieve_chunks()
```
—— 规范引文（DOI、页码）
```
papi ask "..."
```
—— 完整RAG综合分析

MCP Tool Selection (when papi CLI insufficient)

MCP工具选择（当papi CLI功能不足时）

Tool	Speed	Output	Best For
`leann_search(index_name, query, top_k)`	Fast	Snippets + file paths	Exploration, finding which paper to dig into
`retrieve_chunks(query, index_name, k)`	Slower	Chunks + formal citations	Verification, citing specific claims
`papi ask "..."`	Slowest	Synthesized answer	Cross-paper questions, "what does literature say"

Check available indexes:
```
leann_list()
```
or
```
list_pqa_indexes()
```
Embedding priority: Voyage AI → Google/Gemini → OpenAI → Ollama

工具	速度	输出	最佳适用场景
`leann_search(index_name, query, top_k)`	快	片段+文件路径	探索研究，确定需深入研究的论文
`retrieve_chunks(query, index_name, k)`	较慢	文本块+规范引文	验证内容，引用特定论点
`papi ask "..."`	最慢	综合分析结果	跨论文问题，“文献对此有何论述”

查看可用索引：
```
leann_list()
```
或
```
list_pqa_indexes()
```
嵌入模型优先级：Voyage AI → Google/Gemini → OpenAI → Ollama

Adding Papers

添加论文

bash

papi add 2303.13476                   # arXiv ID
papi add https://arxiv.org/abs/...    # URL
papi add 2303.13476 1706.03762 "Attention Is All You Need"  # multiple at once (mixed sources OK)
papi add --pdf /path/to.pdf           # local PDF
papi add --pdf "https://..."          # PDF from URL
papi add --from-file papers.bib       # bulk import

bash

papi add 2303.13476                   # arXiv ID
papi add https://arxiv.org/abs/...    # URL
papi add 2303.13476 1706.03762 "Attention Is All You Need"  # 批量添加（支持混合来源）
papi add --pdf /path/to.pdf           # 本地PDF文件
papi add --pdf "https://..."          # 在线PDF文件
papi add --from-file papers.bib       # 从文件批量导入

Per-Paper Files

单论文文件

Located at

{db}/papers/{name}/

equations.md

summary.md

source.tex

notes.md

paper.pdf

figures/

If agent can't read

~/.paperpipe/

, export to repo:

papi export <papers...> --level equations --to ./paper-context/

Use

--figures

to include extracted figures in export.

See

references/commands.md

for full command reference and per-file details.

存储位置为

{db}/papers/{name}/

：包含

equations.md

、

summary.md

、

source.tex

、

notes.md

、

paper.pdf

、

figures/

。

若Agent无法读取

~/.paperpipe/

，可导出至仓库：

papi export <papers...> --level equations --to ./paper-context/

使用

--figures

参数可在导出时包含提取的图表。

完整命令参考及文件详情请查看

references/commands.md

。