A Swiss Army knife for academics who speak LLM

What it does

GPT Academic is a browser-based control panel for large language models. It bundles dozens of task-specific plugins—paper translation, LaTeX proofreading, code project analysis, and multi-model chat—into one modular interface. You can point it at an arXiv URL, a local PDF, or an entire Python project tree and let the model chew through it.

The interesting bit

The project treats its own codebase as a first-class citizen: it can generate a self-analysis report (self_analysis.md) by feeding its own files back to an LLM. It also renders mathematical formulas in both TeX source and rendered form side-by-side, which is rarer than it should be in chat UIs. The plugin system is hot-reloadable, and the UI buttons are dynamically generated from functional.py, so adding a custom shortcut is just editing a Python file.

Key highlights

• Supports a wide zoo of models—OpenAI GPT, Anthropic Claude, Qwen, GLM, DeepseekCoder, LLaMA2, and more—with parallel querying across several at once.
• Academic-specific plugins: arXiv/LaTeX/PDF translation, Grammarly-style LaTeX proofreading, Google Scholar “related works” generation, and formula rendering.
• Code-intelligence plugins: one-click analysis of Python, C++, Java, or Lua project trees, plus auto-generation of batch function comments.
• Modular, hot-reloadable plugin architecture (crazy_functions) and a “void terminal” that dispatches other plugins via natural language.
• Can run fully offline via local models like ChatGLM3, or mix local and cloud APIs with multiple co-existing keys.

Caveats

• The repository’s main documentation and default UI are Chinese-first; internationalised READMEs are auto-generated by the project itself.
• The standalone void-terminal pip package and multi_language.py script are both marked as experimental or in-development.
• Dependency versions are strictly pinned in requirements.txt, implying a narrow supported environment.

Verdict

Ideal for researchers, students, or developers who need to grind through papers and code with LLMs but don’t want to cobble together ten separate tools. If you already live inside Cursor or a heavily customised Jupyter workflow, this may feel like a parallel universe.