PaperSprint

PaperSprint is a local AI PDF paper reader. It loads papers in the browser, adds AI-guided reading support, and keeps a paper-aware chat beside the PDF.

Features

• Local browser PDF reader with multi-PDF upload, a compact paper list, dark mode, PDF zoom, and collapsible left/right panels.
• AI text analysis with overview, beginner-friendly background notes, takeaways, glossary terms, and guided highlights.
• Guided highlight categories for problem, solution, novelty, method, benchmarking, result, ablation, hyperparams, tradeoff, limitation, and failure. Categories appear only when useful evidence exists.
• PDF highlight interactions: click a highlight to jump/select it, read the short AI comment, explain it in chat, or remove it.
• Manual highlighting: select PDF text to copy, explain in chat, add a highlight, reuse an existing category, or create a new category color.
• Citation parsing and grounding during Analyze for author-year citations, numeric citations, and citation ranges. Clicking a citation opens the full reference-list entry and can add it to chat.
• Takeaways and summary panels that stay beside the PDF while you read.
• Paper-aware chat with optional web search, citation context, figure/table context, reset-and-send, Enter to send, and Shift+Enter for a new line.
• Figure and table analysis after text analysis. Visuals are annotated directly in the PDF as results arrive and can be opened, moved, and added to chat.
• Provider controls for Codex subscription, OpenAI API, and OpenRouter API, with separate text/vision model and reasoning-effort selectors.
• Local cache for parsed papers, analyses, citations, figures, and figure images. Refresh reloads papers and results from cache, and deletion removes the paper from the library and cache.

Quick Start

git clone git@github.com:dimostzim/PaperSprint.git
cd PaperSprint
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m uvicorn app.main:app --host 127.0.0.1 --port 8788

Open http://127.0.0.1:8788.

PaperSprint can use a logged-in Codex CLI subscription by default. API keys are optional:

export OPENAI_API_KEY="sk-..."        # optional
export OPENROUTER_API_KEY="sk-or-..." # optional

Data And Cache

• Current-session PDFs and figure images live under data/session.
• Cached paper records, PDFs, and figure images live under data/cache.
• The refresh button reloads papers and analysis results from data/cache into the current session.
• If a session PDF is missing but the cached PDF exists, PaperSprint restores it automatically.
• Deleting a paper removes it from the visible library and cache.
• Clearing data/cache forces papers to be parsed and analyzed again.