LocalGPT is an open-source application that lets you chat with your documents entirely on your own device using GPT-style models. It has gathered more than 22,000 stars on GitHub, remains actively maintained with updates in August 2026, and is released under the MIT license.
Privacy is the whole point of LocalGPT. Documents are ingested and embedded locally, questions are answered by a locally running language model, and nothing ever leaves your machine, so it works fully offline. Under the hood it builds a retrieval-augmented generation pipeline: your files are split, indexed in a local vector store, and the most relevant passages are fed to the model at question time. Answers come with the source passages they were based on, which makes the tool practical for serious document work.
Getting started is deliberately simple: point LocalGPT at a folder of documents, let it chunk and embed them into a local vector database, then start asking questions. Because the entire pipeline, from embeddings to the language model itself, runs on your own hardware, it can run on a laptop without an internet connection and without any API keys. That makes it a practical fit for air-gapped environments and regulated industries. Users can swap models and embedding providers as better open options appear, and the project community continues to add support for new document types and local inference backends.
LocalGPT is completely free and open source under the MIT license. Your only cost is the local compute needed to run the models.
LocalGPT is ideal for lawyers, researchers, analysts, and anyone handling sensitive documents who wants AI question-answering without sending data to a third-party cloud service.