QAnything

QAnything is a self-hosted knowledge base for people and teams who want to ask questions about their own documents, including collections that mix Chinese and English. It can answer in either language regardless of the document's language, and runs locally through Docker on Windows, macOS and Linux.

It supports PDFs, Word documents, PowerPoint presentations, Excel spreadsheets, email, Markdown, CSV, text files and images, plus web pages. Its document parsing handles tables that span pages, text in multiple columns and embedded images. Answers can include images, and source previews let you check the material behind a response. You can also inspect and edit the extracted text used for retrieval.

Search combines keyword matching with semantic retrieval, then uses BCEmbedding to rerank results before answering. That approach helps find relevant passages in large document collections, including across Chinese and English. A retrieval-only mode returns search results without calling a language model. For smaller tasks, you can ask about an uploaded file without first creating a knowledge base, or use chat without documents. Custom bots and an API support other uses.

The document processing and retrieval services run on CPU, with a stated minimum of 20 GB RAM. QAnything doesn't bundle a local LLM: it connects to Ollama or another OpenAI-compatible API. With a local model, it supports fully offline use and keeps processing on your hardware; choosing a cloud model sends requests to that service. Web search requires internet access. The project is open source under AGPL-3.0.