Academic Paper Analyst renders each page of an academic PDF to an image and sends it to a vision-language model, so figures, tables, and layout are actually seen - not just the extracted text. You get a structured report with page citations, then chat with the paper. Runs on a local model or in the cloud on Claude.
A vision model looks at every page, so nothing that lives in a figure or a table is lost.
Every page is rendered to an image with PyMuPDF and read by a vision-language model - figures, tables, equations, and layout, not OCR'd text.
Overview, contributions, methodology, findings, figure analysis, strengths, limitations, and a TL;DR - a consistent format every time.
The analyzed pages attach as a clickable Source side panel, so every claim can be checked against the paper itself.
Ask follow-up questions about the same document - the model keeps the paper in context and answers in the chat.
Connect filesystem or fetch MCP servers from the UI to load PDFs from disk or pull in web context, called through the tool interface.
Default is a local model in LM Studio - your paper never leaves the machine. Set ANTHROPIC_API_KEY to run it in the cloud on Claude.
The path a paper travels, from upload to a report you can question.
Drop a paper PDF into the chat (page and size limits guard the token budget).
PyMuPDF renders each page to an image and downscales past a max edge.
The pages go to a vision model as images, in order, with the analysis prompt.
A structured, page-cited analysis streams back into the chat.
Follow up on the same paper, or connect MCP tools for more context.
A Chainlit chat app over an OpenAI-compatible client, pointed at a local or cloud vision model. The UI is styled to match the multimodal RAG AI tutor.
Fully offline with LM Studio, or one Blueprint away from a cloud deploy on Render with Claude.
Local runs stay on your machine. A cloud deploy sends page images to Claude's API; gate it with a password for shared use.