Private deployment for shared Zotero libraries — your index, your hardware, your models. Built for teams with 10,000+ documents and rules about where data may live.
One engine, from a personal library to a 10,000-document corpus. Measured on a MacBook Pro (Intel Core i9), macOS 14.3.1 — not on a server.
The index and the retrieval run inside your network. Nothing about your library leaves it unless you point PapersGPT at an external model.
Built-in local LLMs, Ollama, or any OpenAI-compatible endpoint (vLLM, LM Studio, llama.cpp) on your own hardware. Fully offline if you need it.
Researchers keep their Zotero workflow — PDFs, colour-coded annotations, tags and standalone notes all stay in place.
Each researcher runs PapersGPT locally against their own synced Zotero library. No server needed.
One machine indexes the shared library once; the whole team queries the same index.
Index, retrieval and model all run on a machine with no internet access.
No. Indexing and retrieval are local. Content only leaves the machine if you configure an external model — and you can point PapersGPT at an internal inference server instead.
Yes. Built-in local models, Ollama and self-hosted OpenAI-compatible endpoints all work with no internet access.
About 10,000 PDFs / 42 GB of index in 7 minutes on a laptop, with ~20 ms average retrieval. See the scalability report.
Running this for a lab, a department or a company library? We scope private deployments: your own index, your own models, annual license.
