TextGen (text-generation-webui)
Desktop app for local LLMs — text, vision, tool-calling, 100% private
About
TextGen — formerly known as text-generation-webui (the old oobabooga/text-generation-webui repository now redirects to oobabooga/textgen) — is an open-source desktop app for running large language models locally. It pairs a chat UI (instruct, chat, and character modes with Jinja2 prompt templates, message editing, and conversation branching) with chat/completions backends including llama.cpp, ik_llama.cpp, Transformers, ExLlamaV3, and TensorRT-LLM.
Beyond chat it offers vision/multimodal input, file attachments, tool-calling (including MCP servers and web search), an OpenAI/Anthropic-compatible local API, LoRA fine-tuning, and a diffusers-based image generation tab. The README states it is 100% offline with zero telemetry, external resources, or remote update requests. Licensed AGPL-3.0.
What it can do
- • Desktop app + web UI for local LLMs: instruct, chat-instruct, and chat modes with automatic Jinja2 prompt formatting
- • Multiple backends switchable without restart: llama.cpp, ik_llama.cpp, Transformers, ExLlamaV3, TensorRT-LLM
- • Vision: attach images for multimodal understanding; file attachments (txt, PDF, docx)
- • Tool-calling: custom .py tools (web search, page fetch, math) and MCP server support
- • OpenAI/Anthropic-compatible local API (Chat, Completions, Messages endpoints) — drop-in replacement for cloud APIs
- • Training tab: fine-tune LoRAs on chat or raw-text datasets with resume support
- • Image generation tab for diffusers models (e.g. Z-Image-Turbo) with 4/8-bit quantization and a persistent gallery
- • Conversation tools: message editing, version navigation, branching; notebook tab for free-form generation
- • Privacy posture per README: 100% offline, zero telemetry, no external resources or remote update checks
Setup
- Fastest path: download a portable build from https://github.com/oobabooga/textgen/releases, unzip, and double-click 'textgen' — CUDA, Vulkan, ROCm, and CPU-only builds are provided with all dependencies included
- Or use the one-click installer: git clone https://github.com/oobabooga/textgen, then run start_linux.sh / start_macos.sh / start_windows.bat and pick your GPU vendor when prompted (uses Miniforge conda under the hood; ~10GB disk for the full install)
- Download a GGUF model from Hugging Face and drop it into user_data/models (the UI detects it automatically); Transformers/EXL3 multi-file models go in their own subfolder and need the full install
- Start the UI and open http://127.0.0.1:7860 in your browser (desktop builds open their own window)
- To expose the local API (OpenAI/Anthropic-compatible with tool-calling), pass the --api flag or add it to user_data/CMD_FLAGS.txt
- Optional: install extensions (TTS, voice input, translation) via the update wizard script for your OS
Prerequisites
A GGUF (or Transformers/EXL3) model file from Hugging Face; GPU recommended for larger models, CPU-only builds work for small quants; Python 3.9+ only for the manual venv install path
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 47,730 GitHub stars (log-scaled vs catalog peers); last push 56d ago; 30 releases/12mo; 850 open issues. Manual review: rebranded TextGen (oobabooga/textgen); 100% offline, zero telemetry. License: AGPL-3.0.
★ 47,730 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.7/10
- Maintenance · recency, releases, issue load
- 8.3/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 47,730 GitHub stars (log-scaled vs catalog peers)
- · last push 56d ago; 30 releases/12mo; 850 open issues
- · manual review: rebranded TextGen (oobabooga/textgen); 100% offline, zero telemetry
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.