Read research papers by exploring their methods, testing their equations, and comparing their results.
简体中文 · Install · Quick start · Demos · npm
PaperMaster is a local reading workspace for research papers. Import a PDF or an arXiv link, then follow a method from its architecture diagram to an interactive equation and the experiments that test it. Save your findings as Markdown.
Requires Node.js 22.16 or later. The npm package includes the built interface.
npm install -g papermaster-reader
papermasterPaperMaster opens your browser and prints its local URL. In Windows PowerShell, use npm.cmd and papermaster.cmd if script execution policy blocks the shims.
- Open 模型连接 (Model connections) and add a provider. For a Codex subscription, install the Codex CLI and run
codex loginwith your ChatGPT account. You can select a model from the CLI's list. - Click 导入论文 (Import paper), then upload a text-based PDF or paste an arXiv URL.
- Choose 完整精读 (Deep reading). Follow the interactive diagrams in the text, try the formula scenarios, and compare results in the experiment chapter. 泛读 (Skim) includes an animated overview. Select a repository snapshot to ground the diagrams in the available code.
- Save an observation to the main note or export the reading as Markdown. On an older reading, 生成交互版 creates a new interactive record from the same paper version.
Both recordings use reading results for Visual Scaffolding: Toward Native Multimodal Reasoning.
Skim the architecture. Play the flow inside the article and watch the diagram and the selected module explanation advance together.
Try a formula in context. Switch between scenarios and see which grid cells are selected when their scores cross the threshold.
| Feature | Workflow |
|---|---|
| Architecture explorer | Play or step through a method inline. Select a module to inspect its input, operation, and output. |
| Formula lab | Try preset scenarios or edit inputs. Deep reading can include separate labs for different steps of the method. |
| Experiment comparison | Filter reported results and ablations, compare rows, and open the original figures or tables. |
| Guided reading | Work through questions with feedback and resume where you left off. |
| Repository reading | Connect a public GitHub repository and read the implementation at a recorded commit. Requires Git. |
| Markdown notes | Save experiment comparisons or a specific formula run, and export results for later use. |
Connect a Codex subscription, an OpenAI-compatible endpoint, or a native OpenAI, Claude, or Gemini API. API connections use an environment variable or the Windows/macOS system credential store.
The paper library, reading history, and notes stay on your machine. Generation sends the needed paper excerpts to your selected model provider. The reading form lets you choose whether to include figure and table crops. Import currently requires a PDF with extractable text. Interactive coverage depends on the material available in the paper.
git clone https://github.com/Kstheme/PaperMaster.git
cd PaperMaster
npm ci
npm test
npm run build
npm startUse papermaster --help for local server options. See the interactive reading design and platform design for implementation details. Report reproducible bugs at GitHub Issues.

