> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentkey.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Academic paper search for AI agents

> Discover arXiv papers, read paper content, query PDFs and code repos, and search AMiner papers with details, citations and recommendations for AI agents.

Literature work is slow for an agent because the useful parts — ranking what
is new on a topic, reading a paper's actual content, checking a number in the
PDF, opening the code — each live behind a different interface. Through
AgentKey your agent does all of it with `find_tools`, `describe_tool`,
`execute_tool`, with alphaXiv behind the scenes for arXiv papers and their
content, and AMiner for a broad scholarly index with citations and
recommendations.

## What you can ask

* *"What are the newest papers on speculative decoding for LLM inference?"*
* *"Summarize the method and results of 'Attention Is All You Need'."*
* *"In this paper's PDF, which hyperparameters did they use for pretraining?"*
* *"Show me the repository structure of this paper's code and read its README."*
* *"Find NeurIPS papers on graph neural networks by authors at Tsinghua."*
* *"What does this paper cite, and how many times has it been cited?"*
* *"Recommend recent papers for someone interested in protein structure prediction."*
* *"搜索标题包含'大语言模型'的论文，并给出 DOI"*

## What's covered

**Discovery**

* Rank candidate papers for a research topic — literature discovery, related
  work, broad coverage, or what is newest
* Search papers by title, keyword, abstract, author, institution or venue;
  title-only search returning title, id and DOI; results carry a
  citation-count bucket
* Recommendations from a set of topic words or from a scholar's profile
* Papers of a venue (filterable by year), of an institution, or of a
  researcher

**Reading a paper**

* Paper content as text — a structured report built for LLM consumption by
  default, falling back to the full extracted text
* Filtered page content from the PDF for a specific question, addressed by
  paper id, URL or title, so you need not locate the paper first
* Files and directories from the paper's code repository, starting with the
  repository structure and top-level files

**Paper metadata and citations**

* Paper detail: title, authors with affiliations, DOI, keywords, venue, year
  and abstract, with Chinese title and abstract where available
* Batch information for a list of paper ids
* The reference list a paper cites and its citation count

Google Scholar results are a search capability and live under
[Scholar search](/capabilities/search/scholar-search).

## How to use it

Pass the request in plain language and let the router pick the tool:

```text theme={null}
find_tools(q="newest papers on speculative decoding for LLM inference")
find_tools(q="根据标题搜索论文并获取 DOI")
find_tools(prefix="academic/alphaxiv")       # arXiv discovery, content, PDF queries, code, metadata, references
find_tools(prefix="academic/aminer")         # paper search, detail, citations, recommendations
describe_tool("<returned Provider/Operation>")
execute_tool("<Provider/Operation>", { ...params })
```

Both English and Chinese phrasing work. Always `describe_tool` before
executing — AMiner detail tools take an AMiner paper id from a search, while
alphaXiv accepts an id, URL or title directly.

> [!NOTE]
>
> **Building with code?** The same capability is available as fixed-contract
> [Scenario Methods](/api-reference/scenario/overview) with automatic provider
> failover:
> [`academic.paper.search`](/api-reference/scenario/methods/academic-paper-search),
> [`academic.paper.get`](/api-reference/scenario/methods/academic-paper-get).
> `academic.paper.get` takes an arXiv id or arxiv.org URL (served by
> alphaXiv's public API: title, abstract, authors, first submission date and
> links) or the AMiner paper id that `academic.paper.search` returns (which
> adds DOI, venue and citation count). Reading a paper's content or asking
> questions of its PDF stays with the catalog's alphaXiv tools.

## FAQ

> **Which corpus am I searching?**
>
> The alphaXiv route covers arXiv papers and is the one that returns a
> paper's content. The AMiner route is a broad scholarly index across
> venues, with bilingual metadata. Google Scholar is reached through the
> Search category.

> **Can I get the full text?**
>
> For arXiv papers, yes: request the paper's content as text, or ask a
> question of the PDF and get back only the relevant pages. AMiner returns
> metadata, abstracts and citation data rather than the body text.

> **What identifiers do I need?**
>
> AMiner detail tools take AMiner paper ids, which the search tools return —
> start with a search, then open the detail. alphaXiv's agent tools accept a
> paper id, an arXiv URL or the paper title; its REST tools
> (`alphaxiv-api/*` — metadata, citation count, references) take a bare
> arXiv id.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.