AnythingLLM

Building a private ChatGPT usually means gluing a vector store, a document pipeline and a permissions layer onto whichever model you picked. AnythingLLM ships all of that as one application that runs locally by default and needs no setup work first.

You connect a local or cloud model, ingest documents and start chatting within minutes. Support spans llama.cpp compatible models, OpenAI, Anthropic, Azure, Bedrock, Gemini, Ollama, LM Studio, Mistral and Groq among many others, alongside embedders, transcription models and vector databases.

Several features push it past a plain chat wrapper.

• Dynamic model routing: sends each conversation to the provider and model your rules select.
• Managed memories: the model retains information about you or a workspace across sessions.
• Scheduled tasks: recurring prompts run on a cron schedule with full agent capabilities.
• Skill selection: enables unlimited tools while cutting token use per query substantially.
• Agent builder: assembles custom agents and flows without code, with MCP compatibility.

Multi-user permissioning and the embeddable website chat widget are Docker-only, while desktop builds cover Mac, Windows and Linux. A developer API is available for custom integrations, and the team is separately working on Open Computer, an environment giving agents a whole machine to use.