# TextGen

> Source: https://aiwiki.ai/wiki/text_generation_webui
> Updated: 2026-11-18
> Fact-checked: 2026-11-18
> Categories: AI Inference, AI Tools & Products, Developer Tools, Open Source AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "TextGen." aiwiki.ai, 18 Nov 2026. https://aiwiki.ai/wiki/text_generation_webui
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**TextGen** (formerly **text-generation-webui**, also written Text Generation Web UI) is an open-source application for running [large language models](https://aiwiki.ai/wiki/large_language_model) on a user's own computer. Its author and maintainer publishes under the handle oobabooga, and the software is often called by that name.[1][2] The project started in December 2022 as a [Gradio](https://aiwiki.ai/wiki/gradio) web interface for text generation models and was renamed TextGen in April 2026; its README now describes it as "A desktop app for local LLMs. Open source, no telemetry."[3][4][5]

The program is a front end and server that sits on top of separate inference libraries. As of November 2026 it loads models through [llama.cpp](https://aiwiki.ai/wiki/llama_cpp), ik_llama.cpp, Hugging Face [Transformers](https://aiwiki.ai/wiki/transformers_library), ExLlamaV3 and NVIDIA's [TensorRT-LLM](https://aiwiki.ai/wiki/tensorrt_llm), and it offers a chat interface, a notebook for free-form completion, tool calling, [LoRA](https://aiwiki.ai/wiki/lora) training, an image generation tab and an API that follows the OpenAI and Anthropic request formats.[5] The code is licensed under the GNU Affero General Public License version 3, a [copyleft](https://aiwiki.ai/wiki/copyleft) license approved by the Open Source Initiative.[6][7]

On 19 November 2026 the [GitHub](https://aiwiki.ai/wiki/github) repository had 47,738 stars and 5,976 forks.[8] The most recent release at that date was v4.9, published on 2026-05-20.[9]

## History

### Origins as a Gradio web UI

The repository was created on 2022-12-21 under the name text-generation-webui.[8][10] The README of the following day described it as "A gradio webui for running large language models locally. Supports gpt-j-6B, gpt-neox-20b, opt, galactica, and many others", and stated the project's aim: "Its goal is to become the AUTOMATIC1111/stable-diffusion-webui of text generation."[3] The reference is to the [AUTOMATIC1111](https://aiwiki.ai/wiki/automatic1111) web interface for [Stable Diffusion](https://aiwiki.ai/wiki/stable_diffusion), which the current README still credits as the inspiration.[5] Models were downloaded from [Hugging Face](https://aiwiki.ai/wiki/hugging_face), installation used a Conda environment named `textgen`, and the interface was served at port 7860 after running `python server.py`.[3]

In March 2023 a user opened an issue proposing a shorter name such as "textgen". The author declined, replying: "This project is inspired by stable-diffusion-webui, hence the name."[11]

### 2023: LLaMA, new loaders and a grant

Support for other model families and runtimes arrived quickly in early 2023. A loader for [RWKV](https://aiwiki.ai/wiki/rwkv) models was merged on 2023-03-01, a LoRA training tab on 2023-03-28, and llama.cpp support, contributed through a pull request, on 2023-03-31.[12][13][14] One-click installers were first published as a release dated 2023-02-09, whose page lists archives for Windows, Linux, macOS and WSL, and were maintained in a separate repository until they were merged into the main one in September 2023.[15][16]

Tom's Hardware used the program in March 2023 for an article on running a [ChatGPT](https://aiwiki.ai/wiki/chatgpt)-like model on a home PC. Jarred Walton ran [LLaMA](https://aiwiki.ai/wiki/llama) models in 4-bit [GPTQ](https://aiwiki.ai/wiki/gptq) form on a range of Nvidia graphics cards and published benchmarks, and wrote that "Getting the webui running wasn't quite as simple as we had hoped, in part due to how fast everything is moving within the LLM space."[1]

An ExLlama loader followed on 2023-06-16 and ExLlamaV2 loaders on 2023-09-12.[17][18] The first numbered release, v1.0, was published on 2023-07-11.[19] After v1.7 in October 2023 the project announced it was "Switching to a rolling release model with weekly snapshots", and weekly snapshot releases ran from 2023-10-15 to 2024-04-28.[20][9] Numbered releases resumed with v1.8 in June 2024.[21]

In August 2023 [Andreessen Horowitz](https://aiwiki.ai/wiki/andreessen_horowitz) announced an Open Source AI Grant program and named oobabooga in its first group of recipients, describing the funding as a grant and not an investment. The other recipients in that group included the developers of [vLLM](https://aiwiki.ai/wiki/vllm).[22] The README acknowledges the grant.[5]

### 2024 and 2025: redesign and portable builds

Version 1.8 (June 2024) added TensorRT-LLM as a backend.[21] Version 2.0 (December 2024) restyled the interface after the early-2023 look of ChatGPT, moved navigation and past chats into sidebars, and improved use on phones; the release notes credit the chatbot-ui project for much of the CSS.[23]

In April 2025 three releases changed how the program is built and shipped. Version 2.7 added ExLlamaV3 support, version 2.8 replaced the llama-cpp-python binding with a loader that drives llama.cpp's `llama-server` program, and version 3.0 introduced portable builds: self-contained archives of about 700 MB with llama.cpp included that need no installation. Version 3.0 also made llama.cpp the default loader.[24][25][26] Later 3.x releases added file attachments and web search (v3.4, May 2025), image input for vision models in the interface and API (v3.10, August 2025) and an image generation tab based on [Diffusers](https://aiwiki.ai/wiki/diffusers) (v3.20, December 2025).[27][28][29]

### Version 4 and the rename to TextGen

Version 4.0, released on 2026-03-07, replaced the stock Gradio library with the author's own fork, rewrote the training code, reworked tool calling in the API and removed the ExLlamaV2 loader, whose upstream project the notes describe as archived.[30] The releases of the following weeks added tool calling in the chat interface, an Anthropic-compatible endpoint, the ik_llama.cpp backend and support for [Model Context Protocol](https://aiwiki.ai/wiki/model_context_protocol) servers.[31][32][33][34]

The rename took place in mid-April 2026. On 2026-04-13 the README title was changed from "Text Generation Web UI" to "TextGen", and a commit titled "Rename project from text-generation-webui to textgen" followed the same day.[35][36] That commit touched 39 files: it replaced the old repository address in the README, documentation, Dockerfiles, build workflows, the Colab notebook and the installer script, and changed the name the program shows in its window title, log messages and command-line help from "Text Generation Web UI" to "TextGen".[36] The GitHub repository moved from `oobabooga/text-generation-webui` to `oobabooga/textgen`, and requests for the old address resolve to the new one.[8] The release notes for the 4.5 series, published on 2026-04-15, announced the change with the line "The project has been renamed to TextGen!" and gave the new address.[4]

Neither the commit nor the release notes states a reason for the rename.[36][4] The short name was not new inside the project: the Conda environment had been called `textgen` in the README since December 2022, and the portable archives had been named `textgen-portable` since version 3.0.[3][26] The same commit pointed the README's link to the directory of community extensions at its new name, `oobabooga/textgen-extensions`.[36][37]

The README's one-line description changed during the same period. On 13 April it was changed from "A Gradio web UI for running Large Language Models locally" to "The original local LLM interface."[38] Version 4.7, released on 2026-05-03, turned the portable builds into a desktop application: in the words of the notes, "Portable builds now bundle Electron and open as a native window."[39] The README now leads with the desktop app and presents the browser-based web UI as the result of the manual and full installation routes.[5]

### Selected releases

| Version | Date | Notable changes |
|---|---|---|
| v1.0 | 2023-07-11 | First numbered release [19] |
| v1.8 | 2024-06-27 | Numbered releases resume; TensorRT-LLM backend [21] |
| v2.0 | 2024-12-19 | Redesigned interface [23] |
| v2.7 | 2025-04-09 | ExLlamaV3 support [24] |
| v2.8 | 2025-04-18 | llama.cpp loader rebuilt on `llama-server` [25] |
| v3.0 | 2025-04-22 | Portable builds; llama.cpp becomes the default loader [26] |
| v3.4.1 | 2025-05-31 | File attachments, web search (first shipped in v3.4, 2025-05-29) [27][9] |
| v3.10 | 2025-08-12 | Image input in the interface and API [28] |
| v3.20 | 2025-12-07 | Image generation tab [29] |
| v4.0 | 2026-03-07 | Custom Gradio fork, rewritten training, ExLlamaV2 loader removed [30] |
| v4.1.1 | 2026-03-18 | Tool calling in the chat interface (first shipped in v4.1, 2026-03-16) [31][9] |
| v4.2 | 2026-03-24 | Anthropic-compatible `/v1/messages` endpoint [32] |
| v4.3.3 | 2026-04-04 | ik_llama.cpp backend (first shipped in v4.3, 2026-04-03) [33][9] |
| v4.4 | 2026-04-07 | Remote MCP servers [34] |
| v4.5.2 | 2026-04-15 | Rename to TextGen [4] |
| v4.6.2 | 2026-04-23 | Tool call confirmation, local (stdio) MCP servers [40] |
| v4.7.3 | 2026-05-03 | Electron desktop window, interface overhaul [39] |
| v4.8 | 2026-05-07 | Redesigned chat input, `--no-electron` flag [41] |
| v4.9 | 2026-05-20 | Linux ARM64 builds; cross-origin API access limited to localhost by default [42] |

### Activity since mid-2026

Version 4.9, published on 2026-05-20, was the latest release on 19 November 2026.[9] The last commits to the main branch that changed program files are dated 2026-06-01 (an interface fix and an update of the bundled llama.cpp version), and the most recent commit of any kind, on 2026-08-17, edited the README to remove a sponsor banner.[43][44] On 2026-08-02 a user opened a thread on the repository's discussion board asking whether the project was still active; the maintainer had not replied in it by 19 November 2026.[45] The repository was not archived on that date.[8]

## Features

### Chat, instruct and notebook modes

The Chat tab has three modes. In `chat` mode the prompt is a plain transcript built from a character's name, a context string and the conversation so far, which the documentation recommends only for base models. In `instruct` mode the conversation is formatted with the instruction template the model was trained on, which the documentation calls "an offline ChatGPT". `chat-instruct` mode wraps a character conversation inside a single instruction turn so that instruction-tuned models can play a character.[47] Templates are Jinja2 files and are detected from the model's metadata where possible.[5][47]

Messages can be edited, regenerated and continued, earlier versions of a message can be browsed, and a conversation can be branched at any message.[5][47] The Default and Notebook tabs take a raw prompt with no chat structure; they also show the most likely next tokens with their probabilities and the token IDs of a prompt.[48] Text files, PDF and .docx documents can be attached to a message, and models with vision support accept images.[5]

### Tool calling, web search and MCP

Since the 4.1 releases of March 2026 a model can call functions during a chat. Each tool is a single Python file in `user_data/tools` containing a description and an `execute` function, and five are included: `web_search`, `fetch_webpage`, `calculate`, `get_datetime` and `roll_dice`.[31][49] The web search tool queries DuckDuckGo.[49] Remote [MCP](https://aiwiki.ai/wiki/model_context_protocol) servers can be listed in the chat sidebar, local ones that communicate over standard input and output are configured in `user_data/mcp.json`, and an optional setting asks the user to approve each tool call before it runs.[49][40] Over the API, tools are passed in the OpenAI `tools` format for [function calling](https://aiwiki.ai/wiki/function_calling), and the client runs them and returns the results.[50]

### Training and image generation

The Training tab [fine-tunes](https://aiwiki.ai/wiki/fine_tuning) LoRA adapters on a model loaded with the Transformers loader, from chat datasets or raw text, and can resume an interrupted run from a checkpoint.[51][5] The training code was rewritten for version 4.0, whose notes say it was brought in line with [Axolotl](https://aiwiki.ai/wiki/axolotl).[30] A separate tab generates images with Diffusers models and keeps a gallery of results with their settings.[5][29]

### Extensions

An extension is a folder containing a `script.py` that defines hook functions, for example to modify the input or output text or add interface elements, and is enabled with the `--extensions` flag.[52] The documentation lists 13 built-in extensions, covering [text-to-speech](https://aiwiki.ai/wiki/text_to_speech_ai) (Silero and Coqui XTTS), microphone input through [Whisper](https://aiwiki.ai/wiki/whisper), translation through Google Translate, [retrieval-augmented generation](https://aiwiki.ai/wiki/retrieval_augmented_generation) over documents (superbooga and superboogav2), and image requests sent to an AUTOMATIC1111 server.[52][53] Community extensions are listed in a separate directory repository created in April 2023.[52][37]

### Privacy claims

The README states that the program is "100% offline and private, with zero telemetry, external resources, or remote update requests."[5] Some optional features reach beyond the machine when the user turns them on: the web search tool, the Google Translate extension, the `--share` option, which creates a public URL for the interface, and the `--public-api` option, which creates one for the API through [Cloudflare](https://aiwiki.ai/wiki/cloudflare).[49][52][5]

## Model loaders

A loader is the component that reads a model file and runs it. The program selects one from the model's format, or the user picks it with `--loader`.[5]

### Current loaders

The table reflects the documentation for the current release line (4.x) as read in November 2026.[54][55]

| Loader | Model format | Loads LoRAs | Trains LoRAs | Image input |
|---|---|---|---|---|
| llama.cpp | [GGUF](https://aiwiki.ai/wiki/gguf) | No | No | Yes, with an mmproj file |
| Transformers | Full-precision weights, or bitsandbytes 4-bit and 8-bit | Yes | Yes | Through the `send_pictures` extension |
| ExLlamav3_HF | EXL3 | No | No | No |
| ExLlamav3 | EXL3 | No | No | Yes |
| TensorRT-LLM | TensorRT-LLM engines for NVIDIA GPUs | No | No | No |

The llama.cpp loader starts a `llama-server` process and passes it options such as the number of layers to place on the GPU, the [context size](https://aiwiki.ai/wiki/context_window), [KV cache](https://aiwiki.ai/wiki/kv_cache) [quantization](https://aiwiki.ai/wiki/quantization) and the multi-GPU split mode.[54][25][30] The `--ik` flag, or the separate `textgen-portable-ik` builds, substitute ik_llama.cpp, a fork of llama.cpp that the release notes describe as adding new quantization types.[33][5] ExLlamav3_HF pairs the ExLlamaV3 engine with the samplers of the Transformers library, while the plain ExLlamav3 loader uses the engine's own samplers and supports [speculative decoding](https://aiwiki.ai/wiki/speculative_decoding) with a draft model.[54] The older GGML file format no longer loads.[54]

Portable builds contain only the llama.cpp loader and therefore run only GGUF models. Transformers and ExLlamaV3 need the full installation.[5]

### Earlier loaders

The set of loaders has changed many times. The dates below come from the repository's commits and release notes.

| Loader or library | Added | Removed or replaced |
|---|---|---|
| RWKV | 2023-03-01 [12] | Removed in the snapshot of 2023-12-31 [56] |
| llama.cpp through Python bindings | 2023-03-31 with the `llamacpp` package; llama-cpp-python from 2023-04-06 [14][46] | Replaced by the `llama-server` loader in v2.8, 2025-04-18 [25] |
| FlexGen | Before v1.5 | Removed in v1.5, 2023-07-26 [57] |
| ExLlama (version 1) | 2023-06-16 [17] | Removed in the snapshot of 2023-12-31 [56] |
| llamacpp_HF | v1.2, 2023-07-16 [58] | Removed in v2.8, 2025-04-18 [25] |
| ctransformers | v1.6, 2023-09-22 [16] | Removed in the snapshot of 2024-04-07 [59] |
| ExLlamaV2 | 2023-09-12 [18] | Removed in v4.0, 2026-03-07 [30] |
| AutoAWQ | v1.7, 2023-10-08 [60] | Dropped as a standalone loader in v1.12, 2024-07-25, and from the requirements in v1.15, 2024-10-01 [61][62] |
| AutoGPTQ | Present by v1.3, 2023-07-18 [63] | Dropped from the requirements in v1.15; loader removed in v2.2, 2025-01-09 [62][64] |
| ExLlamaV3 | v2.7, 2025-04-09 [24] | Current |
| TensorRT-LLM | v1.8, 2024-06-27 [21] | Current |
| ik_llama.cpp | v4.3 series, April 2026 [33] | Current |

The version 4.0 notes advise users of [EXL2](https://aiwiki.ai/wiki/exl2) models to move to EXL3.[30]

## API

Starting the program with `--api` enables an HTTP server on port 5000 that the documentation presents as a drop-in replacement for the OpenAI and Anthropic APIs.[50] It implements the chat completions, completions, embeddings, models, image generation and moderation routes of the [OpenAI-compatible API](https://aiwiki.ai/wiki/openai_compatible_api) convention, with streaming and tool calls, plus internal routes for listing and loading models and for reading token probabilities.[50] [Embeddings](https://aiwiki.ai/wiki/embeddings) are produced by a [Sentence Transformers](https://aiwiki.ai/wiki/sentence_transformers) model, not by the loaded language model.[50]

The API began as an extension, `extensions/openai`. It became the project's default API in November 2023, after which SillyTavern's documentation referred to the earlier interface as the "Legacy API", and in version 4.2 the code moved out of the extensions folder into the core program.[65][2][32] Version 4.2 also added a `/v1/messages` endpoint in Anthropic's format; the release notes give `ANTHROPIC_BASE_URL=http://127.0.0.1:5000 claude` as the way to point [Claude Code](https://aiwiki.ai/wiki/claude_code) at a local model.[32] Since version 4.0 the llama.cpp, ExLlamaV3 and TensorRT-LLM loaders can serve several requests in parallel.[30]

An API key can be set with `--api-key`, a separate key for administrative actions such as loading models with `--admin-key`, and `--nowebui` runs the API without the interface.[5]

## Installation and distribution

The README describes three routes.[5]

**Portable builds.** Archives for Windows, Linux and macOS have been attached to releases since version 3.0,[26] in variants for CUDA, Vulkan, ROCm and CPU-only use. The user unpacks one and runs `textgen`. The v4.9 set added Linux ARM64 builds aimed at the [NVIDIA DGX Spark](https://aiwiki.ai/wiki/nvidia_dgx_spark).[5][42] Settings, models and chats live in a `user_data` folder that can be moved to a new version.[42]

**Manual portable install.** The README gives these commands for any Python 3.9 or later:[5]

```bash
# Clone repository
git clone https://github.com/oobabooga/textgen
cd textgen

# Create virtual environment
python -m venv venv

# Activate virtual environment
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

# Install dependencies (choose appropriate file under requirements/portable for your hardware)
pip install -r requirements/portable/requirements.txt --upgrade

# Launch server (basic command)
python server.py --portable --api --auto-launch

# When done working, deactivate
deactivate
```

**Full installation.** This route is needed for the Transformers and ExLlamaV3 loaders, training, image generation and extensions. The user runs `start_windows.bat`, `start_linux.sh` or `start_macos.sh`, chooses a GPU vendor when prompted, and opens `http://127.0.0.1:7860` in a browser. The script sets up a Conda environment with Miniforge inside the program folder; the README says this route needs about 10 GB of disk space and downloads [PyTorch](https://aiwiki.ai/wiki/pytorch).[5] Docker Compose files are provided for NVIDIA, AMD, Intel and CPU-only systems.[5]

Setup steps for other local runners are in [How to run LLMs locally](https://aiwiki.ai/wiki/how_to_run_llms_locally).

## License and funding

The repository's LICENSE file is the GNU Affero General Public License, version 3, and the file has not been changed since the initial commit.[6][66] The Open Source Initiative lists AGPL-3.0 among its approved licenses.[7] Section 13 of the license requires anyone who modifies the program and lets users interact with the modified version over a network to offer those users its source code.[6] The license applies to the application's code, not to the models a user loads. llama.cpp, which the portable builds bundle, is published under the MIT license.[5][67] AI Wiki's article on [open-source licenses](https://aiwiki.ai/wiki/open_source_license) covers the differences between these terms.

The repository belongs to a personal GitHub account, not an organization.[8] The README is written in the first person and credits the 2023 Andreessen Horowitz grant as support for the author's "independent work on this project".[5] Release notes from 2024 invited donations through GitHub Sponsors and Ko-fi.[21] The README carried a sponsorship banner for Warp from October 2025 until it was removed in August 2026.[68][44]

## Adoption

GitHub's API reported the following for the repository on 19 November 2026.[8][69][9]

| Measure | Value |
|---|---|
| Stars | 47,738 |
| Forks | 5,976 |
| Watchers | 349 |
| Contributor accounts listed | 380 |
| Releases published | 113 |
| Repository created | 2022-12-21 |
| Latest release | v4.9, 2026-05-20 |

Abubakar Abid of the Gradio team, writing on the Hugging Face blog in April 2025, named the project (under its former name) with AUTOMATIC1111 among "some of the most popular open-source projects of all time" that use Gradio.[70]

## Security

The National Vulnerability Database holds nine CVE records for the program, published between October 2025 and April 2026.[71]

| CVE | Published | Issue | Fixed in |
|---|---|---|---|
| CVE-2025-62364 | 2025-10-13 | Arbitrary file read through a symbolic link in the character picture upload [72] | 3.14 |
| CVE-2025-12487 | 2025-11-06 | Remote code execution without authentication through the `trust_remote_code` parameter sent to the join endpoint; CVSS 9.8 [73][78] | 3.15, through a commit of 2025-10-15 [75][84] |
| CVE-2025-12488 | 2025-11-06 | The same flaw through the load endpoint; CVSS 9.8 [74][83] | 4.0 according to the project's advisory; ZDI and NVD cite the commit of 2025-10-15 [85][83][74] |
| CVE-2026-35050 | 2026-04-06 | Saving extension settings could overwrite Python files of the program, leading to code execution; CVSS 9.1 from GitHub, 8.8 from NVD [76] | 4.1.1 |
| CVE-2026-35483, 35484, 35485, 35487 | 2026-04-07 | Path traversal in the loading of templates, presets, grammars and prompts, readable without authentication [71][77] | 4.3 |
| CVE-2026-35486 | 2026-04-07 | Server-side request forgery in the superbooga extensions [71][77] | 4.3 |

The two `trust_remote_code` flaws were reported through the Zero Day Initiative (ZDI). Its advisories say the vulnerabilities were reported to the developer on 2025-03-12 (CVE-2025-12488) and 2025-03-13 (CVE-2025-12487) and that both advisories were released on 2025-10-30; both name the same commit of 2025-10-15 as the correction.[83][78] That commit made the `--trust-remote-code` setting unchangeable from the interface and the API.[75] The project's own advisories, published on 2026-03-18, give different fix versions for the two flaws: versions up to 3.14 affected and 3.15 patched for CVE-2025-12487 (ZDI-CAN-26681), but versions up to 3.22 affected and 4.0 patched for CVE-2025-12488 (ZDI-CAN-26680).[84][85] Apart from these two, the repository's advisory page lists three further advisories from March and April 2026 that have no CVE number, among them a file write and delete flaw rated critical.[77]

Later releases tightened defaults. Version 4.0 blocked the `extra_flags` and `trust_remote_code` parameters when a model is loaded over the API and restricted file writes to the user data directory, and version 4.9 limited cross-origin API access to localhost unless the user opts into network exposure with `--listen` or `--public-api`.[30][42] The documentation warns that models requiring `trust-remote-code` run Python files shipped with the model, which "can potentially include malicious code".[54]

## Relation to other local LLM software

TextGen is one of several programs for [local AI](https://aiwiki.ai/wiki/local_ai) that put an interface and an API on top of an inference engine.[5] The table gives the code license and star count of related projects on 19 November 2026, from GitHub's API.

| Project | Code license | GitHub stars |
|---|---|---|
| TextGen | AGPL-3.0 [8] | 47,738 [8] |
| [llama.cpp](https://aiwiki.ai/wiki/llama_cpp) | MIT [67] | 135,646 [67] |
| [Ollama](https://aiwiki.ai/wiki/ollama) | MIT [79] | 187,564 [79] |
| [KoboldCpp](https://aiwiki.ai/wiki/koboldcpp) | AGPL-3.0 [80] | 12,346 [80] |
| SillyTavern | AGPL-3.0 [81] | 35,803 [81] |

TextGen depends on **llama.cpp**: its default loader runs llama.cpp's server program.[25][26] **SillyTavern** is a chat front end that connects to a separate model API. Its documentation lists "Oobabooga TextGeneration WebUI" among the local options beside KoboldCpp, llama.cpp, Ollama and TabbyAPI, describes it as an "All-in-one Gradio UI with streaming" with the "Broadest support for quantized (AWQ, Exl2, GGML, GGUF, GPTQ) and FP16 models", and adds a caution: "Regular updates, which can sometimes break compatibility with SillyTavern".[2] The list of formats in that description no longer matches the current loaders: the AWQ, GPTQ and EXL2 loaders have been removed and GGML files no longer load, as recorded above. The same page characterizes **KoboldCpp** as a single binary for GGUF models and **Ollama** as the easiest of the llama.cpp-based APIs to set up.[2]

**LM Studio**'s documentation says it runs models with llama.cpp, and with [MLX](https://aiwiki.ai/wiki/mlx) on Apple silicon, and can serve them on OpenAI-like endpoints.[82] Since version 4.7 TextGen's portable builds take the same form of a downloadable desktop application, while its full installation remains the route to backends other than llama.cpp.[39][5]

## See also

- [Local AI](https://aiwiki.ai/wiki/local_ai)
- [How to run LLMs locally](https://aiwiki.ai/wiki/how_to_run_llms_locally)
- [OpenAI-compatible API](https://aiwiki.ai/wiki/openai_compatible_api)
- [llama.cpp](https://aiwiki.ai/wiki/llama_cpp)
- [KoboldCpp](https://aiwiki.ai/wiki/koboldcpp)
- [Ollama](https://aiwiki.ai/wiki/ollama)
- [LM Studio](https://aiwiki.ai/wiki/lmstudio)
- [Jan](https://aiwiki.ai/wiki/jan_ai)
- [Open WebUI](https://aiwiki.ai/wiki/open_webui)
- [Gradio](https://aiwiki.ai/wiki/gradio)

## References

1. Jarred Walton. "How to Run a ChatGPT Alternative on Your Local PC". Tom's Hardware, March 19, 2023. https://www.tomshardware.com/news/running-your-own-chatbot-on-a-single-gpu
2. SillyTavern. "API Connections". SillyTavern Documentation, retrieved November 19, 2026. https://docs.sillytavern.app/usage/api-connections/
3. oobabooga. "text-generation-webui" (README, revision of December 22, 2022). oobabooga/textgen, GitHub. https://github.com/oobabooga/textgen/blob/a90097e0b8be13f68d6c9896591d38c29f01afa7/README.md
4. oobabooga. "v4.5.2" (release notes). oobabooga/textgen, GitHub, April 15, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.5.2
5. oobabooga. "TextGen" (README). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/README.md
6. oobabooga. "LICENSE" (GNU Affero General Public License, Version 3). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/LICENSE
7. Open Source Initiative. "GNU Affero General Public License version 3". opensource.org, retrieved November 19, 2026. https://opensource.org/license/agpl-3.0
8. GitHub. "oobabooga/textgen" (repository metadata; requests for oobabooga/text-generation-webui return this repository). GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/oobabooga/textgen
9. GitHub. "Releases: oobabooga/textgen". GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/releases
10. oobabooga/textgen. "Initial commit" (commit ad71774). GitHub, December 21, 2022. https://github.com/oobabooga/textgen/commit/ad71774a24ffd484232310406197ee3e34f9ecbb
11. tensiondriven. "Proposal: rename text-generation-webui to textgen or textgen-ui" (issue #653, with reply by oobabooga). oobabooga/textgen, GitHub, March 30, 2023. https://github.com/oobabooga/textgen/issues/653
12. oobabooga/textgen. "Merge pull request #149 from oobabooga/RWKV" (commit f3da6dc). GitHub, March 1, 2023. https://github.com/oobabooga/textgen/commit/f3da6dcc8f0b7b19f4bc8242c165b1d42f1eb88b
13. oobabooga/textgen. "Merge pull request #570 from mcmonkey4eva/add-train-lora-tab" (commit 46f6536). GitHub, March 28, 2023. https://github.com/oobabooga/textgen/commit/46f6536fae8650b4c9f8564c45f4b9f0a46bb8e4
14. oobabooga/textgen. "Add llama.cpp support (#447 from thomasantony/feature/llamacpp)" (commit 6fd70d0). GitHub, March 31, 2023. https://github.com/oobabooga/textgen/commit/6fd70d0032ef6b6612b8efe6405bc347118e5ff2
15. oobabooga. "Windows/Linux/MacOS installers" (release notes). oobabooga/textgen, GitHub, February 9, 2023. https://github.com/oobabooga/textgen/releases/tag/installers
16. oobabooga. "v1.6" (release notes). oobabooga/textgen, GitHub, September 22, 2023. https://github.com/oobabooga/textgen/releases/tag/v1.6
17. oobabooga/textgen. "Add ExLlama support (#2444)" (commit 9f40032). GitHub, June 16, 2023. https://github.com/oobabooga/textgen/commit/9f40032d32165773337e6a6c60de39d3f3beb77d
18. oobabooga/textgen. "Add ExLlamaV2 and ExLlamav2_HF loaders (#3881)" (commit c2a309f). GitHub, September 12, 2023. https://github.com/oobabooga/textgen/commit/c2a309f56e0160b268f0be155fd3e168f1c4bb93
19. oobabooga. "v1.0" (release notes). oobabooga/textgen, GitHub, July 11, 2023. https://github.com/oobabooga/textgen/releases/tag/v1.0
20. oobabooga. "snapshot-2023-10-15" (release notes). oobabooga/textgen, GitHub, October 15, 2023. https://github.com/oobabooga/textgen/releases/tag/snapshot-2023-10-15
21. oobabooga. "v1.8" (release notes). oobabooga/textgen, GitHub, June 27, 2024. https://github.com/oobabooga/textgen/releases/tag/v1.8
22. Matt Bornstein and Rajko Radovanovic. "Supporting the Open Source AI Community". Andreessen Horowitz, August 30, 2023. https://a16z.com/supporting-the-open-source-ai-community/
23. oobabooga. "v2.0" (release notes). oobabooga/textgen, GitHub, December 19, 2024. https://github.com/oobabooga/textgen/releases/tag/v2.0
24. oobabooga. "v2.7" (release notes). oobabooga/textgen, GitHub, April 9, 2025. https://github.com/oobabooga/textgen/releases/tag/v2.7
25. oobabooga. "v2.8" (release notes). oobabooga/textgen, GitHub, April 18, 2025. https://github.com/oobabooga/textgen/releases/tag/v2.8
26. oobabooga. "v3.0" (release notes). oobabooga/textgen, GitHub, April 22, 2025. https://github.com/oobabooga/textgen/releases/tag/v3.0
27. oobabooga. "v3.4.1" (release notes). oobabooga/textgen, GitHub, May 31, 2025. https://github.com/oobabooga/textgen/releases/tag/v3.4.1
28. oobabooga. "v3.10 - Multimodal support!" (release notes). oobabooga/textgen, GitHub, August 12, 2025. https://github.com/oobabooga/textgen/releases/tag/v3.10
29. oobabooga. "v3.20" (release notes). oobabooga/textgen, GitHub, December 7, 2025. https://github.com/oobabooga/textgen/releases/tag/v3.20
30. oobabooga. "v4.0" (release notes). oobabooga/textgen, GitHub, March 7, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.0
31. oobabooga. "v4.1.1" (release notes). oobabooga/textgen, GitHub, March 18, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.1.1
32. oobabooga. "v4.2" (release notes). oobabooga/textgen, GitHub, March 24, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.2
33. oobabooga. "v4.3.3 - Gemma 4 support!" (release notes). oobabooga/textgen, GitHub, April 4, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.3.3
34. oobabooga. "v4.4 - MCP server support!" (release notes). oobabooga/textgen, GitHub, April 7, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.4
35. oobabooga/textgen. "Update README" (commit fae17dc). GitHub, April 13, 2026. https://github.com/oobabooga/textgen/commit/fae17dc5b83d18cc04c7ac060555db7ca51989f6
36. oobabooga/textgen. "Rename project from text-generation-webui to textgen" (commit 65cef2c). GitHub, April 13, 2026. https://github.com/oobabooga/textgen/commit/65cef2c73177fa674ebb8b70a2566e9bac7990a8
37. GitHub. "oobabooga/textgen-extensions" (repository metadata). GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/oobabooga/textgen-extensions
38. oobabooga/textgen. "Update README" (commit 61bfc2f). GitHub, April 13, 2026. https://github.com/oobabooga/textgen/commit/61bfc2ffd56559da96dbde3e0d60fc233ec1a5ac
39. oobabooga. "v4.7.3" (release notes). oobabooga/textgen, GitHub, May 3, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.7.3
40. oobabooga. "v4.6.2" (release notes). oobabooga/textgen, GitHub, April 23, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.6.2
41. oobabooga. "v4.8" (release notes). oobabooga/textgen, GitHub, May 7, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.8
42. oobabooga. "v4.9" (release notes). oobabooga/textgen, GitHub, May 20, 2026. https://github.com/oobabooga/textgen/releases/tag/v4.9
43. GitHub. "Commits: oobabooga/textgen" (main branch). GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/commits/main
44. oobabooga/textgen. "Update README" (commit c93f887). GitHub, August 17, 2026. https://github.com/oobabooga/textgen/commit/c93f8871239550de2ccfe1e95d469aa82616f07e
45. oobabooga/textgen Discussions. "No updates in more than 2 months. Is textgen still alive?" (discussion #7634). GitHub, August 2, 2026. https://github.com/oobabooga/textgen/discussions/7634
46. oobabooga/textgen. "Add new llama.cpp library (2048 context, temperature, etc now work)" (commit 03cb44f). GitHub, April 6, 2023. https://github.com/oobabooga/textgen/commit/03cb44fc8ca24c6458dbd25c31acd1e8cbdfcde2
47. oobabooga. "01 - Chat Tab" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/01%20-%20Chat%20Tab.md
48. oobabooga. "02 - Default and Notebook Tabs" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/02%20-%20Default%20and%20Notebook%20Tabs.md
49. oobabooga. "Tool Calling Tutorial" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/Tool%20Calling%20Tutorial.md
50. oobabooga. "12 - OpenAI API" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/12%20-%20OpenAI%20API.md
51. oobabooga. "05 - Training Tab" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/05%20-%20Training%20Tab.md
52. oobabooga. "07 - Extensions" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/07%20-%20Extensions.md
53. oobabooga. "extensions/whisper_stt/requirements.txt". oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/extensions/whisper_stt/requirements.txt
54. oobabooga. "04 - Model Tab" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/04%20-%20Model%20Tab.md
55. oobabooga. "What Works" (documentation). oobabooga/textgen, GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/blob/main/docs/What%20Works.md
56. oobabooga. "snapshot-2023-12-31" (release notes). oobabooga/textgen, GitHub, December 31, 2023. https://github.com/oobabooga/textgen/releases/tag/snapshot-2023-12-31
57. oobabooga. "v1.5" (release notes). oobabooga/textgen, GitHub, July 26, 2023. https://github.com/oobabooga/textgen/releases/tag/v1.5
58. oobabooga. "v1.2" (release notes). oobabooga/textgen, GitHub, July 16, 2023. https://github.com/oobabooga/textgen/releases/tag/v1.2
59. oobabooga. "snapshot-2024-04-07" (release notes). oobabooga/textgen, GitHub, April 7, 2024. https://github.com/oobabooga/textgen/releases/tag/snapshot-2024-04-07
60. oobabooga. "v1.7" (release notes). oobabooga/textgen, GitHub, October 8, 2023. https://github.com/oobabooga/textgen/releases/tag/v1.7
61. oobabooga. "v1.12" (release notes). oobabooga/textgen, GitHub, July 25, 2024. https://github.com/oobabooga/textgen/releases/tag/v1.12
62. oobabooga. "v1.15" (release notes). oobabooga/textgen, GitHub, October 1, 2024. https://github.com/oobabooga/textgen/releases/tag/v1.15
63. oobabooga. "v1.3" (release notes). oobabooga/textgen, GitHub, July 18, 2023. https://github.com/oobabooga/textgen/releases/tag/v1.3
64. oobabooga. "v2.2" (release notes). oobabooga/textgen, GitHub, January 9, 2025. https://github.com/oobabooga/textgen/releases/tag/v2.2
65. oobabooga. "snapshot-2023-11-12" (release notes). oobabooga/textgen, GitHub, November 12, 2023. https://github.com/oobabooga/textgen/releases/tag/snapshot-2023-11-12
66. GitHub. "History for LICENSE: oobabooga/textgen". GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/commits/main/LICENSE
67. GitHub. "ggml-org/llama.cpp" (repository metadata). GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/ggml-org/llama.cpp
68. oobabooga/textgen. "Add sponsor (Warp) to README <3" (commit 898a3ed). GitHub, October 7, 2025. https://github.com/oobabooga/textgen/commit/898a3ed2feb9750f8998f371ef3af52f36a7cb6e
69. GitHub. "oobabooga/textgen contributors". GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/oobabooga/textgen/contributors
70. Abubakar Abid. "Journey to 1 Million Gradio Users!". Hugging Face Blog, April 4, 2025. https://huggingface.co/blog/gradio-1m
71. National Institute of Standards and Technology. "CVE API, keyword search: text-generation-webui" (ten results: nine for this program, and CVE-2026-61732, which is for another product and only mentions it). National Vulnerability Database, retrieved November 19, 2026. https://services.nvd.nist.gov/rest/json/cves/2.0?keywordSearch=text-generation-webui
72. National Institute of Standards and Technology. "CVE-2025-62364 Detail". National Vulnerability Database, retrieved November 19, 2026. https://nvd.nist.gov/vuln/detail/CVE-2025-62364
73. National Institute of Standards and Technology. "CVE-2025-12487 Detail". National Vulnerability Database, retrieved November 19, 2026. https://nvd.nist.gov/vuln/detail/CVE-2025-12487
74. National Institute of Standards and Technology. "CVE-2025-12488 Detail". National Vulnerability Database, retrieved November 19, 2026. https://nvd.nist.gov/vuln/detail/CVE-2025-12488
75. oobabooga/textgen. "Make --trust-remote-code immutable from the UI/API" (commit b5a6904). GitHub, October 15, 2025. https://github.com/oobabooga/textgen/commit/b5a6904c4ac4049823396090360b6f566f4e4603
76. National Institute of Standards and Technology. "CVE-2026-35050 Detail". National Vulnerability Database, retrieved November 19, 2026. https://nvd.nist.gov/vuln/detail/CVE-2026-35050
77. oobabooga/textgen. "Security Advisories". GitHub, retrieved November 19, 2026. https://github.com/oobabooga/textgen/security/advisories
78. Zero Day Initiative. "ZDI-25-982: oobabooga text-generation-webui trust_remote_code Reliance on Untrusted Inputs Remote Code Execution Vulnerability". zerodayinitiative.com, October 30, 2025. https://www.zerodayinitiative.com/advisories/ZDI-25-982/
79. GitHub. "ollama/ollama" (repository metadata). GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/ollama/ollama
80. GitHub. "LostRuins/koboldcpp" (repository metadata). GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/LostRuins/koboldcpp
81. GitHub. "SillyTavern/SillyTavern" (repository metadata). GitHub REST API, retrieved November 19, 2026. https://api.github.com/repos/SillyTavern/SillyTavern
82. LM Studio. "Welcome to LM Studio Docs!". lmstudio.ai, retrieved November 19, 2026. https://lmstudio.ai/docs/app
83. Zero Day Initiative. "ZDI-25-981: oobabooga text-generation-webui trust_remote_code Reliance on Untrusted Inputs Remote Code Execution Vulnerability". zerodayinitiative.com, October 30, 2025. https://www.zerodayinitiative.com/advisories/ZDI-25-981/
84. oobabooga/textgen. "ZDI-CAN-26681: oobabooga text-generation-webui trust_remote_code Reliance on Untrusted Inputs Remote Code Execution Vulnerability" (security advisory GHSA-6j2r-r6xw-423x). GitHub, March 18, 2026. https://github.com/oobabooga/textgen/security/advisories/GHSA-6j2r-r6xw-423x
85. oobabooga/textgen. "ZDI-CAN-26680: oobabooga text-generation-webui trust_remote_code Reliance on Untrusted Inputs Remote Code Execution Vulnerability" (security advisory GHSA-j53j-fh8f-rpw3). GitHub, March 18, 2026. https://github.com/oobabooga/textgen/security/advisories/GHSA-j53j-fh8f-rpw3

