LitGraph

Graphical literature survey for arXiv. Turn any arXiv paper into an interactive map of the work it cites, related research, and the authors connecting them — then grow the map by expanding any node.

Version 2.0 · Created by Shreyan Kundu

---

Overview

Provide a paper — an arXiv identifier, a title, or a PDF — and LitGraph renders an interactive graph of:

• The seed paper at the centre.
• The references it cites, sized by citation count.
• Similar papers, drawn from the whole Semantic Scholar corpus.
• Every author, linked to each of their papers, so collaborators meet at shared work and prolific researchers emerge as hubs.

Every node opens a detail panel with the abstract, citation count and direct arXiv/PDF links. Any node can be expanded to extend the graph, and multiple papers can be analysed together into a single merged map.

LitGraph targets open-access arXiv papers, so every paper on the map resolves to a freely available full text.

Requirements

• Python 3.10+
• Two packages: requests, pypdf
• An internet connection (the graph library loads from a CDN)
• Optional but recommended: a free Semantic Scholar API key — see Configuration

There is no build step, bundler, container or database.

Installation

pip install -r requirements.txt

Running

python backend/server.py

Then open http://localhost:8000. On Windows, run.bat does the same thing. Stop the server with Ctrl+C.

Route	Purpose
/	The application
/docs	Full documentation (also available as a PDF)
/system	Interactive architecture map
/admin	Administration: API key, defaults, cache, password
/health	Health check

Environment variables

Variable	Default	Purpose
PORT	8000	Port to serve on
S2_API_KEY	—	Semantic Scholar API key (the admin panel takes precedence)
LITGRAPH_DATA_DIR	backend/	Where config.json and the response cache are written

First run

1. Start the server and open http://localhost:8000/admin.
2. Set an administrator password. This protects the API key and cache controls.
3. Paste your Semantic Scholar API key (see below). Alternatively set S2_API_KEY.
4. Open http://localhost:8000 and analyse a paper — try 1706.03762.

Obtaining an API key

LitGraph ships with no credentials. Each operator supplies their own.

Request a free key from the Request an API key form at https://www.semanticscholar.org/product/api. It is issued by email.

LitGraph will run without a key against Semantic Scholar's shared anonymous pool, but that pool is rate-limited across all anonymous users worldwide and is unreliable under load. A key is strongly recommended for anything beyond casual use.

Your key is never committed. It is stored in config.json inside the data directory, which is excluded by .gitignore along with the response cache.

Usage

Enter an arXiv identifier in any of these forms:

1706.03762            1706.03762v7            arXiv:1706.03762
https://arxiv.org/abs/1706.03762             (or a pasted PDF link)

…or a paper title, or upload the PDF. Use ＋ Add to queue several papers into one merged graph.

The control bar sets how much of each source to retrieve and how results are prioritised when there are more matches than your limit:

Control	Range	Notes
References — limit	0–1000	Papers the seed cites
References — priority	Most cited / Newest / Oldest / API order
Similar — limit	0–500	Related work
Similar — priority	Most relevant / Most cited / Newest / Oldest
Authors — per paper	0–50	Authors drawn per paper
Authors — from	Every paper / Seed only
2nd hop	0–25	Also pull the references of your top references

Quick / Deep Research / Complex are presets that load starting values into these controls.

Documentation

Complete documentation is served at /docs while the application is running, and is distributed as frontend/LitGraph-Documentation.pdf. It covers every feature in both plain-English and technical form, the verified API limits, the data pipeline, the HTTP API, and a candid account of the system's constraints.

• DATA-AND-LICENSING.md — data sources, endpoints, rate limits, and the licence terms governing the data. Read this before any commercial deployment.
• CHANGELOG.md — release history.
• /system — interactive architecture map.

Architecture

 Browser (vanilla JS + Cytoscape.js)
        │  fetch / JSON
        ▼
 server.py ······ pure Python standard library, no framework
        │
        ├── graph_builder.py   normalise config · build · expand
        ├── pdf_extract.py     PDF → arXiv id / title
        └── config.py          settings · PBKDF2 auth · sessions
                │
                ▼
        sources.py ──► cache.py (30-day disk cache)
                   ──► Semantic Scholar API ──► arXiv links

Each source is retrieved as a pool, filtered to arXiv, ranked, and then trimmed to the requested limit — in that order, so a limit means what it says. Because the pool size is constant, changing a limit re-slices cached data and costs no API call.

Data and attribution

Bibliographic data is retrieved from the Semantic Scholar Academic Graph API, operated by the Allen Institute for Artificial Intelligence. Use of that API requires attribution to Semantic Scholar, which LitGraph displays in the application; this attribution must not be removed.

Data returned by the API is governed by its own licences, which may include non-commercial terms. See DATA-AND-LICENSING.md.

LitGraph is not affiliated with, endorsed by, or sponsored by the Allen Institute for Artificial Intelligence, Semantic Scholar, or arXiv.

Licence

Copyright © 2026 Shreyan Kundu. All rights reserved. Proprietary — see LICENSE.