Tool
text-generation-webui
oobabooga’s local LLM front end — historically the Gradio “text-generation-webui,” now shipping as TextGen (desktop + browser UI, OpenAI/Anthropic-compatible API). Same project lineage; clone URLs that used text-generation-webui still redirect.
What it is
A local app for running open-weight models with a chat/instruct UI, notebook-style generation, optional vision, tool-calling, and an API. Official README: portable builds (CUDA / Vulkan / ROCm / CPU) for GGUF via llama.cpp; a fuller install unlocks extra backends (Transformers, ExLlamaV3, etc.), training, image generation, and extensions. Claims no telemetry on the project’s own description — still verify your extensions and model sources separately.
What it replaces
It replaces hosted chat UIs when you want prompts and weights on your machine, and replaces “wire Gradio yourself on top of llama.cpp” for people who want loaders, character chat, and an OpenAI-shaped API in one place. It can sit beside or instead of thinner UIs (Ollama + a chat front end) when you need loader knobs, LoRA training tabs, or character modes.
It does not replace VRAM. It does not replace reading model licenses. And it is not a multi-tenant production inference farm — that is closer to vLLM / a real serving stack.
License
The application is under the GNU Affero General Public License v3.0 (AGPL-3.0), verified from the project LICENSE on GitHub (oobabooga/textgen; same SPDX on the redirected text-generation-webui path).
AGPL caveat: if you modify it and let users interact with that modified version over a network, you must offer corresponding source under AGPL. Local personal use is the common case; shipping a modified hosted service is where counsel starts caring. Model licenses are separate from the UI license — same rule as every other local runner.
Where it falls over
- Install surface is wide: portable GGUF path vs full Conda/PyTorch path — wrong choice wastes an evening and a disk.
- VRAM/RAM still set the ceiling; a UI with every sampler knob does not invent memory.
--listen/ share / public API flags turn a private toy into a network endpoint; AGPL obligations can follow modified network use.- Extensions and extra backends increase breakage and dependency churn; portable mode intentionally hides some features.
- Rename noise: docs and blog posts still say “oobabooga” / “text-generation-webui” while the product title is TextGen — bookmark the current GitHub org repo, not a random mirror.
Who it's for
Power users who want a feature-dense local LLM UI/API and are fine managing models, loaders, and AGPL. People who only need “pull and chat” may be happier with Ollama + a simpler front end — different job, not a scoreboard.
Hardware that usually matters
The apps on these pages are free or free-to-use. The recurring cost is usually the machine: VRAM first, then system RAM, then disk for weights. Example Amazon listings skewed toward entry and mid budgets for beginners checking fit — not endorsements; stock and price move constantly.
- GIGABYTE GeForce RTX 3060 Eagle OC 12GB — entry 12 GB VRAM pattern many beginners still use for 7B–13B quantized chat before spending mid-range money.
- GIGABYTE GeForce RTX 4060 Ti AERO OC 16GB — mid 16 GB VRAM path when 12 GB starts feeling tight; still not an honest “70B fully on-GPU” card.
- EVGA GeForce RTX 3090 XC3 Ultra 24GB (Renewed) — older 24 GB class for larger local models / less painful offload; renewed stock and warranties differ from new.
- CORSAIR Vengeance LPX DDR4 32GB (2×16GB) — cheaper system RAM path for older DDR4 boxes when layers spill or you CPU-offload.
- Patriot Viper Venom DDR5 32GB (2×16GB) — DDR5 kit for newer boards; 32 GB is a common “I might spill layers” starting point.
- Samsung 990 EVO Plus 1TB NVMe SSD — entry NVMe for a first model library; step up when pulls fill the drive.
- Samsung 990 PRO 2TB NVMe SSD — default library disk for a mixed set of 7B/13B quants without deleting every Tuesday.
Some links are Amazon Associates. We may earn a commission if you buy.
Verdict
Use text-generation-webui / TextGen when you want a deep local UI+API under AGPL-3.0 and you will own the hardware and the model licenses. Keep it off the public internet unless you meant to. Star counts are not VRAM.