Search scientific papers by what you see, not what you remember. Find figures using natural language — "find the Phillips relation", "how does a planet radius change with distance from its host star?" — across the literature.
- Conda (for the
plot_finderPython 3.12 backend env) - Node.js 18+ (frontend)
- Tesseract OCR (
conda install -n plot_finder -c conda-forge tesseract, orbrew install tesseract)
git clone <repo-url> astro-plot-finder
cd astro-plot-finder
cp .env.example .env
# Edit .env — add your ADS_API_KEY (required)
# Get a token at: https://ui.adsabs.harvard.edu/user/settings/token
# ANTHROPIC_API_KEY is optional — see below
conda create -n plot_finder python=3.12
conda run -n plot_finder pip install -r backend/requirements.txt
cd frontend && npm installmake backend # http://localhost:8000
make frontend # http://localhost:5173ANTHROPIC_API_KEY is optional. Without it, the app still works fully, using
a regex-based query parser and skipping the Claude reranking step. Add your
own key in .env to enable Claude-powered query parsing and result reranking
for more accurate matches.
- Paywalled papers are skipped silently.
- Papers before ~1991 may have no arXiv version.
- Scanned-PDF OCR captions are noisier and flagged with a ⚠ badge.