AnythingLLM is an open-source full-stack application designed to create a private knowledge base from local files. It allows users to interact with documents using various Large Language Models (LLMs) without relying on a direct OpenAI subscription. The tool supports multiple formats, including PDF, DOCX, TXT, web pages, and audio via Whisper.
The application offers two deployment modes: a desktop version for individual use and a Docker version for server-based, multi-user environments. The Docker version supports features like user invitations, access control for workspaces, and embedded chat widgets. It also utilizes a Model Router, introduced in version 1.13.0, to automatically select between different models based on query complexity.
For data processing, the tool uses a RAG (Retrieval-Augmented Generation) approach, splitting documents into chunks and storing them in a vector database. While it defaults to LanceDB, it supports external databases like PGVector, Chroma, and Pinecone. Users can connect to various LLM providers, including local options via Ollama or cloud services through OpenAI-compatible gateways.
Security is a critical aspect of the platform. A vulnerability identified as CVE-2026-48116, which allowed arbitrary command execution within Docker containers, was patched in version 1.13.0. Users are advised to keep their instances updated, with version 1.19.0 being the current reference point at the time of the report.
Integration with third-party gateways, such as the one provided by Bothub, allows users to access models like Claude, GPT, and DeepSeek without needing a VPN or international payment methods. Configuration involves setting the base URL to https://openai.bothub.chat/v1 and selecting the appropriate model ID within the Generic OpenAI provider settings.
