TextGen desktop app drops with zero telemetry and ik_llama.cpp quantization
The open-source LLM frontend formerly known as text-generation-webui now ships as a portable Electron app for Windows, Linux, and macOS, with no telemetry and ik_llama.cpp quant support.
TextGen, the open-source LLM frontend that predates LLaMA and llama.cpp, now ships as a native desktop app with no installation required. Developer oobabooga released portable builds for Windows, Linux, and macOS this week — you download a zip, extract it, and double-click the executable. The entire runtime, including chat histories and settings, lives in a single self-contained folder.
The project started in December 2022 as text-generation-webui and has been a staple in the local LLM community. The new desktop packaging uses Electron, the same framework LM Studio runs on, but with a minimal integration that keeps the UI snappy.
What stands out
- 01Zero telemetry. TextGen makes no outbound requests. LM Studio phones home on every launch with OS details, CPU architecture, app version, and backend choices. TextGen doesn't.
- 02ik_llama.cpp builds. The app ships with ik_llama.cpp, a fork that includes IQ4_KS and IQ5_KS quant types. LM Studio and Ollama ship vanilla llama.cpp only. The ik quants deliver better accuracy at the same bit depth.
- 03Built-in web search. A
ddgsPython library integration lets models search the web via tool-calling (works with Qwen 3.6 and Gemma 4) or as text attachments via an "Activate web search" checkbox. - 04Three tool-calling modes. Single-file.py tools, HTTP MCP servers, and stdio MCP servers. Each tool call can show an approve/reject prompt before execution.
- 05Character mode. Beyond instruction-following chats, you can create custom characters for casual conversation.