What it does

TextGen is a desktop application for running large language models on your own hardware. Portable builds for Linux, Windows and macOS come with CUDA, Vulkan, ROCm and CPU-only options, and the site describes setup as downloading, unzipping and double-clicking. Models in GGUF format are placed in a folder and detected automatically. The interface offers instruction and character chat modes, image attachments for vision models, uploaded text, PDF and .docx files, message editing and branching, and a notebook tab for free-form generation. Models can call tools during chat, including web search, page fetching and math, and MCP servers are supported. Additional backends include llama.cpp, ik_llama.cpp, Transformers, ExLlamaV3 and TensorRT-LLM, and models and backends can be switched without restarting. A local API with chat, completions and messages endpoints can stand in for the OpenAI and Anthropic APIs. The full installation adds LoRA fine-tuning, an image generation tab for diffusers models, and extensions for text-to-speech, voice input and translation. The project states that it works fully offline with zero telemetry and is released under AGPL-3.0. Besides the desktop app, it can run as a web UI in the browser through a Python install, a one-click installer or Docker.

Capabilities

Runs GGUF models locally, with portable builds for Linux, Windows and macOS
Chat with image attachments and uploaded text, PDF and .docx files
Local API compatible with OpenAI and Anthropic clients
Tool-calling with web search, plus MCP server support
LoRA fine-tuning and an image generation tab in the full installation
Offline operation with no telemetry

When to use it

You want to run models privately without a cloud service
Existing OpenAI or Anthropic client code should point at a local model
You want to try different backends and models from one interface
You want to fine-tune LoRAs on your own chat or text data