DeepXiv/deepxiv_sdk

DeepXiv SDK – Quick Overview

What it is – A Python package and command‑line tool that talks to the DeepXiv service, a cloud backend that has already parsed the full text of millions of arXiv papers (and, via the partner 1stAuthor, many legal, clinical‑trial, grant and other “vertical‑domain” datasets). The SDK lets you:

• Ask natural‑language questions and get answers that include real arXiv citations.
• Search and read papers section‑by‑section, so you only pull the part you need.
• Explore people (researcher profiles) and other domains (law, court cases, drug trials, SEC filings, CVEs, etc.) with the same fa commands.

Why it matters – Traditional search APIs give you titles/abstracts; to verify a claim you have to download a whole PDF (tens of thousands of tokens). DeepXiv pre‑processes papers into structured sections and runs an “agentic” LLM over the full bodies, returning concise answers with verifiable citations. This saves token budget and makes automated research pipelines feasible.

Key commands

Verb	Example	What it does
deepxiv ask	deepxiv ask "speedup of speculative decoding"	LLM‑powered answer with arXiv IDs.
deepxiv ask --web	deepxiv ask "which providers support speculative decoding" --web	Same, but searches cached web pages.
deepxiv search / deepxiv paper	deepxiv paper 2512.15176 --section Methods	Find papers, list sections, read a single section.
deepxiv fa …	deepxiv fa search talent "world model" -F career_stage=student	Query 1stAuthor vertical domains (researcher profiles, law, trials, etc.).

Python usage – The Reader class mirrors the CLI:

from deepxiv_sdk import Reader
r = Reader(token="YOUR_KEY")
ans = r.agent_search("speedup of DEER on HumanEval")
section = r.section("2512.15176", "Method")
people = r.fa().search("talent", "speculative decoding", top_k=5)

Installation & limits – pip install deepxiv-sdk. The first run auto‑registers a free token (300 agentic calls per day; 3 trial fa ask calls). Higher quotas are available from the service’s registration page or by contacting the maintainers.

Documentation – Full CLI reference, Python API, and domain‑specific guides are in the repo (USAGE.md, FA.md, examples/).

License – MIT.

---

All details are taken directly from the repository’s README.