// the find
oobabooga/textgen
Open-source desktop app for local LLMs. Text, vision, tool-calling, OpenAI/Anthropic-compatible API. 100% private.
This is the rebrand/evolution of text-generation-webui, a long-running Gradio-based desktop app and API server for running local LLMs (GGUF, EXL3, Transformers, TensorRT-LLM). It targets people who want a self-hosted ChatGPT-style UI with an OpenAI/Anthropic-compatible API, without sending data anywhere.
Genuinely mature multi-backend support (llama.cpp, ik_llama.cpp, ExLlamaV3, Transformers, TensorRT-LLM) that most competing UIs don't bother maintaining. The portable builds with bundled CUDA/ROCm/Vulkan binaries actually solve the real pain point of local LLM setup hell. Tool-calling and MCP support plus an OpenAI/Anthropic-compatible API make it usable as a drop-in backend for other tools, not just a standalone chat window. Command-line flag surface is huge, meaning almost every sampler parameter and hardware quirk is configurable if you're willing to read the docs.
The flag list and installation matrix (conda vs portable vs docker vs full) signal real complexity under the hood; new users will hit friction picking the right requirements file for their GPU. Extensions are a grab-bag of scripts with inconsistent maintenance (some, like superboogav2, bundle their own nltk data and look stale). No mention of test coverage or CI beyond build workflows, so quality control for such a sprawling codebase is unclear. Being a years-old project with a huge accumulated flag surface means backward-compat cruft is baked in, and the constant renaming/rebranding (oobabooga/text-generation-webui -> textgen) suggests churn that could break saved configs or bookmarks.