The data layer agentic search is missing — full paper text, real citations, and the people behind them.

Live system · Full documentation · API docs · Status · · 中文

A CLI and Python SDK over a service that has already read the literature. Papers arrive parsed into sections rather than PDFs, retrieval runs over full bodies rather than abstracts, and an agentic endpoint answers questions with citations that resolve to real arXiv IDs and URLs.

Four things it does:

An agent researching a topic has bad options. Search APIs return ten blue links and abstracts — enough to name a paper, never enough to answer "what speedup does it report on HumanEval". PDFs answer that, but cost 50k tokens each and arrive as a wall of text with no structure to navigate.

DeepXiv removes the tradeoff. Papers are pre-parsed, so an agent can spend 300 tokens on a TLDR to decide whether to spend 5k on the Methods section. Questions that need evidence go to the agentic endpoint, which reads source text and hands back an answer with citations you can check. And because knowing who does the work is half of research, the same token searches scholars — plus law, clinical trials, drugs, grants and more through 1stAuthor's domains.

DeepXiv works closely with 1stAuthor to serve data beyond papers: researchers, statutes, court judgments, clinical trials, drugs, grants, filings, standards and CVEs, all reached through deepxiv fa with the same token. → The deepxiv fa guide

One CLI, twelve domains, the same three verbs: search → read → ask.

Pass --domain (-d). Auto-routing (fa ask without -d) is still in early development and can pick the wrong domain. Name the domain yourself; deepxiv fa domains lists them and deepxiv fa spec DOMAIN shows its filters.

# People: China-based students working on world models
deepxiv fa search talent "world model" -F career_stage=student -F location=北京,上海 -k 20

# Which academicians / fellows cited DeepSeek-R1, and what the positive ones said
deepxiv fa read appraise arxiv:2501.12948 -l brief
deepxiv fa read appraise arxiv:2501.12948 -p stance=positive -p org=CAS,CAE

# Law and cases: statutes plus US Supreme Court judgments
deepxiv fa search law "个人信息保护" -k 5
deepxiv fa search cases "qualified immunity police" -F country=us -F court_level=scotus -k 5

# Trials, filings, vulnerabilities
deepxiv fa search trials "GLP-1 obesity" -F phase=PHASE3 -F overall_status=RECRUITING -k 5
deepxiv fa search filings "CEO resignation" -F market=us -F "filed_year>=2023" -k 5
deepxiv fa search cve "*" -F kev=true -F severity=critical,high --sort epss -k 10

# A cited answer over one domain (agentic quota; SDK token: 3 trial asks)
deepxiv fa ask "个人信息出境需要满足什么条件" -d law
deepxiv fa ask "DeepSeek-R1 得到了哪些院士的正面评价" -d appraise -p paper=arxiv:2501.12948

DeepXiv itself stays focused on agentic data services for academic papers, and will remain free. If you need higher limits on the 1stAuthor domains, see 1stauthor.com.

1stAuthor is in closed beta. Questions or problems? Open an issue.

pip install deepxiv-sdk

deepxiv fa (1stAuthor domains, including talent) needs 1.2.0 or later: pip install -U deepxiv-sdk.

deepxiv auto-registers a token on first use. Agentic commands (ask, fa ask) need a registered key (300 calls/day free); the auto-registered token gets 3 lifetime trial fa ask calls. Get a key at data.rag.ac.cn/register, then:

deepxiv config --token YOUR_REGISTERED_KEY

Every account gets 300 agentic calls/day free, on a pool separate from the general daily limit.

One investigation, start to finish. You've heard speculative decoding got much faster this year and want to know what's real.

1. Ask the literature. Start with the question, not a keyword. The service picks its own tools, reads paper bodies, and cites what it used.

deepxiv ask "what speedup does speculative decoding report on HumanEval in 2025"

DEER reports a 5.54× speedup on HumanEval (with Qwen3-30B-A3B as the target
model), compared to EAGLE-3's 2.41× on the same benchmark [arXiv:2512.15176].

📚 Sources (1 cited, 10 retrieved — use --all-sources for the rest):
  1. [2512.15176] DEER: Draft with Diffusion, Verify with Autoregressive Models

deepxiv ask — a question in, a cited answer streaming out

The answer goes to stdout and sources to stderr, so deepxiv ask "…" > answer.md captures just the answer. Add --effort high when a question spans several papers.

2. Read the paper it cited — in layers. Never load a whole paper to answer a question about one section.

deepxiv paper 2512.15176 --brief              # title, TLDR, keywords, citations — worth reading?
deepxiv paper 2512.15176 --head               # section list + where the tokens are
deepxiv paper 2512.15176 --section Experiment # read only that

Each step costs an order of magnitude more than the last, so you stop as soon as you have your answer. Take section names from --head — papers don't share a common outline. --preview gives ~10k chars; no flag at all gives the full markdown.

3. Widen it into a search. Once you know what you're looking for, filter for the rest.

deepxiv search "speculative decoding" --date-from 2025-01 --min-citations 20 --limit 10

Filters combine with AND — --authors, --orgs, --categories, --venue/--venue-year, dates, citation floors. Stack too many and you'll legitimately get zero results; loosen one.

4. Find the people behind it. A method is worth more when you know whose lab it comes from and what else they've built.

deepxiv fa search talent "speculative decoding" --top-k 5    # by topic
deepxiv fa search talent "Geoffrey Hinton"                    # or by name
deepxiv fa search talent "deep learning" -F "org=University of Toronto" -F "h_index>=50"
deepxiv fa read talent 15023                                  # profile: bio, education, career, top papers, metrics
deepxiv fa read talent 15023 --level full                     # the full researched profile (markdown)

Names (Chinese, pinyin, English) and topics both work. IDs from search feed read; deepxiv fa spec talent lists the filters and what each read level costs. deepxiv talent search|survey still work as deprecated aliases, but IDs from the old index don't carry over.

The same verbs cover 1stAuthor's other domains (deepxiv fa domains lists them). Full guide: FA.md.

deepxiv fa search law "个人信息保护" --top-k 5
deepxiv fa ask "个人信息出境需要满足什么条件" --domain law   # agentic quota; SDK token: 3 trial asks

5. Step off arXiv when the question isn't academic. Licensing, pricing, who shipped what last week — same command, different backend.

deepxiv ask "which inference providers support speculative decoding today" --web
deepxiv ask "NeurIPS 2025 best paper" --web --search-type news

The web backend reads cached page bodies. Pages read in full are marked 📄, snippet-only ones 🔗 — weigh them accordingly.

In Python, the same pipeline is Reader:

from deepxiv_sdk import Reader

reader = Reader(token="YOUR_REGISTERED_KEY")   # Reader takes the token explicitly
answer = reader.agent_search("what speedup does DEER report on HumanEval")["answer"]
method = reader.section("2512.15176", "Method")
people = reader.fa().search("talent", "speculative decoding", top_k=5)

• USAGE.md — full CLI reference, the Python API, streaming, error handling, batching, and the built-in research agent. (中文)
• FA.md: deepxiv fa, the 1stAuthor domains (talent, law, cases, trials, drugs, grants, filings, standards, CVEs, …): commands, filters, prices, errors, Python. (中文)
• skills/deepxiv-cli/SKILL.md — drop-in operating instructions for coding agents. Two worked workflows also ship as skills: trending digest, baseline table.
• examples/ — runnable scripts for each entry point.

Also available: PubMed Central, bioRxiv/medRxiv, trending papers, and per-paper social metrics — see USAGE.md.

If DeepXiv is useful in your work, please cite the technical report:

@article{qian2026deepxiv,
  title   = {DeepXiv-SDK: An Agentic Data Interface for Scientific Literature},
  author  = {Qian, Hongjin and Xia, Ziyi and Liu, Ze and Chen, Jianlyu and
             Luo, Kun and Qin, Minghao and Li, Chaofan and Xiong, Lei and
             Lan, Junwei and Wang, Sen and Liang, Zhengyang and Shao, Yingxia and
             Lian, Defu and Liu, Zheng},
  journal = {arXiv preprint arXiv:2603.00084},
  year    = {2026},
  url     = {https://arxiv.org/abs/2603.00084}
}

MIT — see LICENSE.

• 🐛 Issues: github.com/DeepXiv/deepxiv_sdk/issues
• 📧 Higher limits: email tommy[at]chien.io with your use case