TextGen - oobabooga

TextGen is an open-source desktop app for running local LLMs on Windows, macOS, and Linux.It supports text and multimodal (vision) inputs, file attachments (TXT, PDF, DOCX), image understanding, and a notebook tab for free-form generation.

Multiple backends and model formats are supported, including gguf (llama.cpp), exllamav3, transformers, ik_llama.cpp, and tensorrt-llm, with backend switching without restart.Provides chat and instruction-following modes, jinja2 prompt templates, conversation branching, message editing and versioning for prompt engineering workflows.

Includes an OpenAI/Anthropic-compatible API and tool-calling support for custom functions, web search, page fetching, and MCP server integration.Portable builds available with CUDA, Vulkan, ROCm, and CPU-only options; dependencies are bundled for quick local deployment.

Extensible via extensions, training and image-generation backends, and a desktop UI plus API for developers building local LLM applications.

Key Features

• Cross-platform desktop app for running local LLMs on Windows, macOS, and Linux
• Supports text and multimodal (vision) inputs, image understanding, file attachments (TXT, PDF, DOCX), and a notebook tab for free-form generation
• Multiple backends and model formats (gguf/llama.cpp, exllamav3, transformers, ik_llama.cpp, tensorrt-llm) with backend switching without restart
• Chat and instruction-following modes with jinja2 prompt templates, conversation branching, message editing and versioning
• OpenAI/Anthropic-compatible API and tool-calling support for custom functions, web search, page fetching, and MCP server integration

Use Cases

• Build a private, offline multimodal personal assistant on your laptop that ingests PDFs, images and notes, performs document-aware text generation for summaries and Q&A, and switches model backends dynamically for best speed vs. accuracy
• Create a research and writing workspace that attaches source files and datasets, uses the prompt-engineering tools and conversation branching to iterate on literature reviews or grant drafts, and exports reproducible prompts or API/tool-calls for downstream workflows
• Prototype and test LLM-powered automations locally by hot-switching model backends, defining custom function-calling and tool integrations, and iterating on multimodal prompts and branching dialogs before deploying to production

Who is it for?

• Software developers
• Machine learning engineers
• Prompt engineers
• Startup founders
• Local llm hobbyists