Text Generation WebUI
oobabooga/textgenReviewed by AI
Checked by RepoPicks Radar on
What is Text Generation WebUI?
TextGen (formerly text-generation-webui) is a desktop app for AI language models that run on your machine. Portable packages for Windows, macOS and Linux include everything; you add a GGUF model file to the user_data/models folder and chat. It handles images in messages, PDF and Word attachments, message editing and branching, model tool-calling with web search, a tab for generating pictures, LoRA training, and an OpenAI/Anthropic-compatible API other programs can call. Each system has builds for CPU only, NVIDIA (CUDA), AMD (ROCm) and Vulkan.
Text Generation WebUI is free and open source (AGPL-3.0) and can replace ChatGPT and OpenRouter.
How easy is it to install Text Generation WebUI?
B · installer plus one simple step · Windows
- Windows B
- macOS B
- Linux B
All three systems are B: the README says "Download, unzip, double-click textgen. A window opens", and each release has a portable package for them (for example textgen-portable-4.9-windows-cuda12.4.zip, 936 MB, or textgen-portable-4.9-macos-arm64.tar.gz, 272 MB), but you still have to fetch a GGUF model from Hugging Face and drop it into user_data/models. Extra backends, training and image generation need the full install instead, with Conda, PyTorch and about 10 GB of disk.
Text Generation WebUI on GitHub
- Stars
- 47,734on Nov 10, 2026
- Downloads
- 437,7572,666 this week
- Growth
- +3stars in 7 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Nothing extra
- Disk space
- 1.3 GB–4.0 GB · from the installer
- Accounts or keys
- None
- Cost
- Free · Free under AGPL-3.0, a copyleft licence. The author states that Andreessen Horowitz gave a grant for the work in August 2023.
- License
- AGPL-3.0 (Copyleft)
- Can replace
- ChatGPT Plus, OpenRouter
- Last change on GitHub
- Aug 17, 2026
Text Generation WebUI risk signals (5)
- Attentionflags can put it on the internet
The command-line list includes `--listen` ("Make the web UI reachable from your local network"), `--share` ("Create a public URL") and `--public-api` ("Create a public URL for the API using Cloudflare"). A password is a separate option (`--gradio-auth`, `--api-key`), so a public URL without one leaves your model and chats open to anyone with the link.
AI review
- Attentionoption runs code shipped with a model
The flag `--trust-remote-code` is documented as "Set trust_remote_code=True while loading the model. Necessary for some models." With it, Python code that comes inside a downloaded model runs on your computer, and another flag exists specifically to force a file format that "prevents arbitrary code execution".
AI review
- Infoprivacy claim is conditional
The README says "100% offline and private, with zero telemetry, external resources, or remote update requests." The app still reaches the internet when you use the documented features: you download models from Hugging Face, and the tool-calling built in does "web search, page fetching, and math".
AI review
- Infofull install downloads a lot at run time
The full installation "Requires ~10GB disk space and downloads PyTorch", and the one-click script "uses Miniforge to set up a Conda environment in installer_files/", so most of what runs is fetched after you start the script, not from the release file you downloaded.
AI review
- Infoportable builds lag the code
The newest release is v4.9 from 2026-05-20, while the last commit is from 2026-08-17 — the ready-made packages are several months behind the code, and there have been no commits since August with 867 issues open.
AI review
Evidence is shown in its original language.
What Radar could not check about Text Generation WebUI
- Whether the portable build starts on your particular graphics card, and which of the CUDA, ROCm, Vulkan or CPU packages you need
- Real disk use after downloading models, which are usually several gigabytes each
- Whether the app truly sends nothing out, which can only be seen by watching the network while it runs
- How fast answers are on a computer without a graphics card
- Whether the image generation and training tabs work as described