Key Takeaways
- text-generation-webui (github.com/oobabooga/textgen) is free and open source under AGPL 3.0, with 47,600+ GitHub stars.
- It was created by developer "oobabooga" in December 2022, inspired by AUTOMATIC1111's Stable Diffusion WebUI.
- In April 2026 (version 4.5.2) the project renamed itself "TextGen" and moved to a new repository; the old name and URL still redirect there.
- Supported backends today are llama.cpp, ik_llama.cpp, Transformers, ExLlamaV3, and TensorRT-LLM β older loaders like AutoGPTQ and ExLlamaV2 have been dropped.
- A May 2026 update added a native Electron desktop app, closing much of the setup-complexity gap with turnkey tools like LM Studio.
π In One Sentence
text-generation-webui is a free, open-source (AGPL 3.0) local LLM interface, created by developer "oobabooga" in December 2022, that runs models directly through llama.cpp, ExLlamaV3, Transformers, or TensorRT-LLM, and in April 2026 renamed itself "TextGen" while shipping a native desktop app.
π¬ In Plain Terms
Unlike Open WebUI, text-generation-webui doesn't just talk to a model server β it IS the engine that loads and runs the model on your machine, with a huge number of dials to tune exactly how it generates text.
Quick Facts: text-generation-webui (TextGen)
- License: AGPL 3.0 (fully open source)
- Platforms: Windows, Linux, macOS (portable Electron builds or full Python/Conda install)
- Website / GitHub: github.com/oobabooga/textgen (formerly github.com/oobabooga/text-generation-webui)
- Stats: 47,600+ stars, ~6,000 forks, ~380-395 contributors
- Cost: Free, no company, no paid tier
- Core function: Runs local models directly (llama.cpp, ExLlamaV3, Transformers, TensorRT-LLM) with a configurable chat UI, LoRA training, and a tool-calling API
What Is text-generation-webui?
text-generation-webui is a free, open-source application that both loads and runs a local language model AND provides the chat interface for it β unlike server-fronting tools such as Open WebUI, it is the inference engine itself, not just a client. It supports several different backend engines so you can choose the format your model is distributed in: GGUF via llama.cpp, EXL3 via ExLlamaV3, full-precision via Transformers, or optimized inference via TensorRT-LLM.
It ships as either a portable desktop build (unzip and double-click, with an optional native Electron window) or a full Python installation for users who also want LoRA fine-tuning and image generation, which require a larger dependency set.
Its defining trait since 2022 has been depth of control: sampling parameters, prompt templating, extensions, and an OpenAI/Anthropic-compatible API with tool-calling are all built in, at the cost of a steeper learning curve than single-purpose consumer apps.
History: From oobabooga's Side Project to "TextGen"
The repository was created on December 21, 2022, by a developer known only by their GitHub handle, "oobabooga." Their real name has not been publicly disclosed in any source we could verify β despite the project's prominence, the maintainer has kept their identity private throughout. The project's own documentation describes it as built in the spirit of, and using the same Gradio framework as, AUTOMATIC1111's Stable Diffusion WebUI β the equivalent tool for image generation that predates it.
In August 2023, the maintainer received an open-source grant from venture capital firm Andreessen Horowitz (a16z) to support continued independent, full-time development β one of the earlier examples of a VC firm directly funding an individual open-source maintainer rather than a company. (We could not independently confirm the exact grant amount from a primary a16z source; treat that detail as unconfirmed.)
The project's biggest change arrived in 2026, not as a UI tweak but as a full rebrand: version 4.5.2, released April 15, 2026, renamed the project from "text-generation-webui" to "TextGen" and moved the GitHub repository to `github.com/oobabooga/textgen` (the old URL redirects automatically). A rapid run of releases followed: 4.6.2 (April 23) added tool-call confirmation and stdio MCP server support; 4.7.3 (May 3) shipped a native Electron desktop app β the single biggest usability change in the project's history, turning a browser-tab tool into a real double-click application β plus DGX Spark aarch64 builds and llama.cpp tensor parallelism; 4.8 (May 7) redesigned the chat composer; and 4.9 (May 20) added MTP speculative decoding and a live tokens-per-second display.
Alongside the rebrand, the supported backend list changed materially: AutoGPTQ, AutoAWQ, and the original ExLlama/ExLlamaV2 loaders are no longer part of the current README, replaced by llama.cpp, ik_llama.cpp, Transformers, ExLlamaV3, and TensorRT-LLM as the current lineup.
What Can You Actually Do With text-generation-webui?
- Run a model with the backend that fits your hardware. Pick llama.cpp for GGUF quantized models, ExLlamaV3 for EXL3 quantization on NVIDIA GPUs, Transformers for full-precision, or TensorRT-LLM for optimized NVIDIA inference β all from the same UI.
- Fine-tune a model with LoRA. The Training tab supports LoRA fine-tuning on chat-formatted or raw-text datasets, with pausable and resumable training runs β a feature most turnkey chat apps don't offer at all.
- Automate through a tool-calling API. An OpenAI- and Anthropic-compatible API server supports custom Python functions and MCP (Model Context Protocol) servers, so external scripts and agents can call tools through the same model you're chatting with.
- Switch chat modes for the task at hand. Instruct mode for direct Q&A, Chat mode for persona-based conversation, Chat-Instruct as a hybrid, and Notebook mode for raw free-form text completion outside of a conversation format.
- Attach images and documents. Vision-capable models can process attached images, and PDF/DOCX files can be uploaded directly into a conversation.
- Generate images from the same app. A built-in image-generation tab supports diffusers-based image models (such as Z-Image-Turbo) alongside your text model.
Who Should Use text-generation-webui?
text-generation-webui is built for people who want to see and control every setting affecting how a local model generates text β not for someone who wants an app that makes every choice for them.
- β You want to fine-tune a model with LoRA without a separate training pipeline β the built-in Training tab handles this directly.
- β You want a choice of backend engine per model β llama.cpp, ExLlamaV3, Transformers, and TensorRT-LLM are all available from one app.
- β You want to build agents or scripts against a local tool-calling API β the OpenAI/Anthropic-compatible API with MCP support covers this without a third-party bridge.
- β You want a real desktop app now, not just a browser tab β the 2026 Electron rebuild gives you a double-click launch experience.
- β You want the absolute simplest one-click chat experience with no settings to understand β try LM Studio instead.
- β You need character cards, lorebooks, and roleplay-specific tooling β pair SillyTavern as the frontend with text-generation-webui as one possible backend.
- If unsure: download the portable build for your OS and GPU, double-click it, and load a small GGUF model through llama.cpp first β you can explore LoRA training and the API later.
When Should You NOT Use text-generation-webui?
Depth of control is the whole point of text-generation-webui, and it is also exactly what makes it a poor fit in a few situations.
- If you want zero configuration: even with the 2026 Electron app, there are more settings (sampling parameters, backend choice, prompt templates) than a single-purpose consumer chat app exposes.
- If you need built-in multi-user accounts or team access controls: there is no login system or role-based access β it is a single-user local application. Use Open WebUI if you need that.
- If roleplay with character cards and lorebooks is your main use case: text-generation-webui has no dedicated roleplay tooling; pair it with SillyTavern as the frontend instead.
- If you need a guaranteed support contract: it is a community project maintained by a single developer, with no commercial support tier or SLA.
- If you only ever want cloud models: it is built around running models locally on your own hardware; it is not a client for managing cloud provider API keys the way tools like LibreChat are.
How Do You Install text-generation-webui?
There are two supported install paths depending on whether you need training and image generation:
- 1. Portable build (recommended for most users). Download the archive matching your OS and GPU from github.com/oobabooga/textgen/releases β separate builds exist for CUDA, Vulkan, ROCm, and CPU-only. Unzip, then double-click the `textgen` executable; a native Electron window opens by default (pass `--no-electron` for the classic browser-tab UI at `http://127.0.0.1:7860`).
- 2. Full install (needed for LoRA training and image generation). Clone the repository and run `start_windows.bat`, `start_linux.sh`, or `start_macos.sh`, which sets up a Conda environment and installs dependencies automatically (roughly 10 GB of disk space).
- 3. Docker. Separate Dockerfiles are provided per vendor (NVIDIA, AMD, Intel, CPU); run `docker compose up --build` from the matching directory.
- 4. Download a model. Use the built-in Model tab to pull a model directly from Hugging Face, or manually place a GGUF file in the `models/` folder, then select the matching backend.
- 5. Pick a chat mode. Choose Instruct, Chat, Chat-Instruct, or Notebook mode under the Parameters tab depending on your use case.
- Full documentation is maintained at the project wiki.
Pricing: Free and Open Source
There is no pricing tier to compare β text-generation-webui has always been fully free.
- Free, forever, for everyone. The software is licensed under AGPL 3.0 with no paid version, no subscription, and no company selling it.
- No hosted version. It is designed to run on your own hardware; there is no official cloud-hosted "TextGen as a service" offering.
- Funding history. The maintainer received a one-time Andreessen Horowitz open-source grant in 2023 to support full-time development β this funded the developer's time, not a commercial product, and did not change the license or introduce a paid tier.
- Your only real cost is hardware. Since it runs models locally, the practical cost is whatever GPU or CPU you already own β there is no per-token or per-message fee from the software itself.
text-generation-webui vs KoboldCpp vs LM Studio vs Open WebUI: Which One?
These four solve overlapping but distinct problems β three run models directly, one is a front end for a separate backend.
Tool | Link | Best for | Runs models directly | LoRA training | Multi-user | License |
|---|---|---|---|---|---|---|
| text-generation-webui (TextGen) | github.com/oobabooga/textgen | Deep configurability: multiple backends, LoRA training, tool-calling API | Yes (llama.cpp, ExLlamaV3, Transformers, TensorRT-LLM) | Yes | No | AGPL 3.0 |
| KoboldCpp | github.com/LostRuins/koboldcpp | Lightweight, creative-writing/roleplay-friendly backend, easy SillyTavern pairing | Yes (llama.cpp-based) | No | No | AGPL 3.0 |
| LM Studio | lmstudio.ai | Simplest turnkey GUI for downloading and running GGUF models | Yes (native) | No | No | Free (closed source) |
| Open WebUI | openwebui.com | Self-hosted, multi-user front end for a separately run backend | No (needs Ollama/vLLM/API) | No | Yes, with roles | Open WebUI License |
For character/roleplay-focused chat that can use text-generation-webui as a backend, see SillyTavern. For a broader backend-vs-serving-engine comparison, see text-generation-webui vs vLLM vs llama.cpp.
Common Mistakes With text-generation-webui
- Loading a GGUF file with the wrong backend selected. Match the backend to the model format: llama.cpp for GGUF, ExLlamaV3 for EXL3, Transformers for full-precision safetensors.
- Searching GitHub for "text-generation-webui" and assuming it's abandoned because the top result is now "TextGen." It is the same actively maintained project under a new name and repository URL as of April 2026 β the old URL simply redirects.
- Using an outdated guide that references AutoGPTQ or the original ExLlama. Those loaders are no longer part of the current backend lineup; use llama.cpp or ExLlamaV3 instead.
- Expecting built-in multi-user login. There is no account system β if multiple people need separate access, that requires a different tool such as Open WebUI in front of a shared backend.
- Trying to use it as a roleplay frontend with character cards. It has no lorebook or character-card system; pair it with SillyTavern for that workflow instead.
Common Questions About text-generation-webui
Is text-generation-webui the same thing as TextGen?
Yes. In version 4.5.2 (April 15, 2026) the project renamed itself from "text-generation-webui" to "TextGen" and moved its GitHub repository to github.com/oobabooga/textgen. It is maintained by the same developer and carries forward the same star count and commit history; the old repository URL redirects to the new one.
Who created text-generation-webui?
A developer who goes by the handle "oobabooga" created the project, with the first commit dated December 21, 2022. Their real name has never been publicly disclosed in any source we could verify.
Is text-generation-webui free?
Yes, completely. It is licensed under AGPL 3.0, has no company behind it, and no paid tier. The maintainer received a one-time Andreessen Horowitz open-source grant in 2023 to support full-time development, but that did not create a commercial product.
What backends does text-generation-webui support?
As of 2026, the supported backends are llama.cpp, ik_llama.cpp, Transformers, ExLlamaV3, and TensorRT-LLM. Older backends from earlier versions β AutoGPTQ, AutoAWQ, and the original ExLlama/ExLlamaV2 β are no longer listed as supported in the current codebase.
Does text-generation-webui support LoRA fine-tuning?
Yes. The Training tab supports LoRA fine-tuning on either multi-turn chat datasets or raw text, and training runs can be paused and resumed.
Does text-generation-webui have an API?
Yes. It exposes both an OpenAI-compatible and an Anthropic-compatible API (covering Chat, Completions, and Messages-style endpoints), with tool-calling support through custom Python functions or MCP servers, and optional --api-key/--admin-key authentication.
Is text-generation-webui good for roleplay?
It can be used for roleplay thanks to its detailed sampling-parameter controls, but it is not roleplay-first software β it has no character cards, lorebooks, or group chat. For a roleplay-specific frontend, see SillyTavern, which can use text-generation-webui as one of its backends.
Does text-generation-webui need a browser now?
Not necessarily. Since version 4.7.3 (May 2026), the portable builds open as a native Electron desktop window by default. Pass the --no-electron flag if you prefer the classic browser-tab interface at http://127.0.0.1:7860.
How is text-generation-webui different from KoboldCpp?
KoboldCpp is purpose-built for creative writing and roleplay and is commonly used as a lightweight drop-in backend for SillyTavern. text-generation-webui is broader and more configurable, with more backend engines, LoRA training, and a fuller API, at the cost of more settings to understand.
How is text-generation-webui different from Open WebUI?
Open WebUI is a server-based, multi-user front end that typically sits in front of Ollama or another backend. text-generation-webui is a single-user desktop/local app that is itself the inference engine host β it runs the model directly rather than connecting to a separately managed one.
Sources
- text-generation-webui / TextGen source code, README, and releases β github.com/oobabooga/textgen
- GitHub API repository data (creation date, license, star count)
- Project wiki and documentation β github.com/oobabooga/textgen/wiki