Generate anything — text, images, video. Locally. Uncensored.
No cloud. No data collection. No API keys. Auto-detects 12 local backends. Your AI, your rules.
The only desktop app that runs AI chat, image, and video generation — locally, one click, no cloud.
Download · Features · Quick Start · Why This App? · Roadmap
| Chat with Personas | Image / Video Generation |
|---|---|
 |
 |
| Model Manager | Create View with Parameters |
 |
 |
GLM-4.7-Flash, Model Loading Fix, Agent Badge Audit, 75+ Downloadable Models
- GLM-4.7-Flash — ZhipuAI's strongest 30B class model. 11 variants across uncensored (Heretic) and mainstream, IQ2 to Q8. Fits 12GB VRAM (IQ2_M). Native tool calling.
- GLM 5.1 754B MoE — Frontier agentic engineering model listed as cloud-available via Ollama.
- Model Loading Fix — Fixed 3 bugs causing "0 models loaded" in ComfyUI Create View (race condition at startup, broken auto-retry logic, stale cache after download). Models now load reliably within seconds.
- Agent Badge Audit — Consistent agent flags across all 75+ models. Models with native tool calling are correctly marked.
- Removed HOT Badges — Cleaner UI, only AGENT badges shown for tool-calling models.
- ComfyUI Plug & Play — Auto-detect, one-click install, auto-start. Zero config image and video generation.
- 20 Model Bundles — 8 image + 12 video bundles with one-click download.
- Z-Image Turbo/Base — Uncensored image model. 8-15 seconds per image. No safety filters.
- FLUX 2 Klein — Next-gen FLUX architecture with Qwen 3 text encoder.
- Image-to-Image (I2I) — Upload a source image, adjust denoise strength, transform with any image model.
- Image-to-Video (I2V) — FramePack F1, CogVideoX, SVD with drag & drop image upload.
- Dynamic Workflow Builder — 14 strategies. Auto-detects installed nodes.
- Unified Download Manager — Track all downloads with progress, speed, retry.
- Think Mode in Chat Input — Toggle thinking mode directly from the message input area.
- Process Cleanup — ComfyUI auto-terminates when app is closed (Windows Job Object).
| Feature | Locally Uncensored | Open WebUI | LM Studio | SillyTavern |
|---|---|---|---|---|
| AI Chat | Yes | Yes | Yes | Yes |
| Coding Agent (Codex) | Yes | No | No | No |
| 13 MCP Agent Tools | Yes | No | No | No |
| Plug & Play Setup | 12 Backends | No | Built-in | No |
| Multi-Provider (20+ Presets) | Yes | Yes | Yes | No |
| A/B Model Compare | Yes | No | No | No |
| Local Benchmark | Yes | No | No | No |
| Image Generation | Yes | No | No | No |
| Image-to-Image | Yes | No | No | No |
| Image-to-Video | Yes | No | No | No |
| Video Generation | Yes | No | No | No |
| File Upload + Vision | Yes | Yes | Yes | No |
| Thinking Mode | Yes | No | No | No |
| Granular Permissions | 7 Categories | No | No | No |
| Uncensored by Default | Yes | No | No | Partial |
| Memory System | Yes | Plugin | No | No |
| Agent Workflows | Yes | No | No | No |
| Document Chat (RAG) | Yes | Yes | No | No |
| Voice (STT + TTS) | Yes | Partial | No | No |
| Open Source | AGPL-3.0 | MIT | No | AGPL |
| No Docker | Yes | No | Yes | Yes |
- Plug & Play Setup — First-launch wizard auto-detects 12 local backends. Nothing installed? One-click in-app Ollama download and install with progress bar. ComfyUI one-click install with step-by-step progress. Configurable ComfyUI port and path in Settings. Zero config needed.
- Uncensored AI Chat — Abliterated models with zero restrictions. Streaming + thinking display.
- Multi-Provider — 20+ presets. Local: Ollama, LM Studio, vLLM, KoboldCpp, llama.cpp, LocalAI, Jan, TabbyAPI, GPT4All, Aphrodite, SGLang, TGI. Cloud: OpenAI, Anthropic, OpenRouter, Groq, Together, DeepSeek, Mistral. Switch per conversation.
- Codex Coding Agent — Reads codebase, writes code, runs shell commands. File tree with native folder picker. Up to 20 tool iterations.
- Agent Mode — 13 MCP tools: web search, file I/O, shell, code execution, screenshots, system info. Native + Hermes XML fallback.
- Image Generation — FLUX 2 Klein, FLUX.1 (schnell/dev), Z-Image Turbo/Base, Juggernaut XL, RealVisXL, DreamShaper XL via ComfyUI. Full parameter control, no content filter.
- Image-to-Image — Upload a source image, adjust denoise strength, transform with any image model.
- Video Generation — Wan 2.1, HunyuanVideo 1.5, LTX 2.3, AnimateDiff Lightning, CogVideoX, FramePack F1 on your GPU.
- Image-to-Video — FramePack F1 (6 GB VRAM), CogVideoX 5B, SVD-XT. Upload an image, get video.
- Thinking Mode — Provider-agnostic. See the AI's reasoning before the answer. Toggle from chat input.
- File Upload + Vision — Drag & drop, paste, clip button. Vision models analyze images.
- Granular Permissions — 7 tool categories, 3 permission levels, per-conversation overrides.
- Smart Tool Selection — Reduces tool definitions per request by ~80%. JSON repair for local LLMs.
- Memory System — Persistent across conversations. Auto-extraction. Export/import.
- Agent Workflows — Multi-step chains. 3 built-in (Research, Summarize URL, Code Review). Visual builder.
- Model A/B Compare — Same prompt, two models, side by side. Parallel streaming.
- Local Benchmark — One-click benchmark any model. Tokens/sec leaderboard.
- Document Chat (RAG) — Upload PDFs, DOCX, TXT. Hybrid search with source citations.
- Voice Chat — Push-to-talk STT + sentence-level TTS streaming.
- 20+ Personas — Pre-built characters. Switch without prompt engineering.
- Chat Export — Markdown or JSON. Token counter. Keyboard shortcuts.
- Standalone Desktop App — Tauri v2 Rust backend. Download .exe, run it.
- Model Load/Unload — Power icons in header. Load into VRAM, unload when done.
- Custom Dark Titlebar — Frameless window, no native chrome.
- Linear/Arc UI — Compact, monochrome. 15% larger for readability.
- Privacy First — Zero tracking, all API calls proxied locally.
- Desktop: Tauri v2 (Rust backend, standalone .exe)
- Frontend: React 19, TypeScript, Tailwind CSS 4, Framer Motion
- State: Zustand with localStorage persistence
- AI Backend: 20+ providers (Ollama, LM Studio, vLLM, KoboldCpp, llama.cpp, LocalAI, Jan, OpenAI, Anthropic, OpenRouter, Groq, and more), ComfyUI, faster-whisper
- Build: Vite 8 (dev), Tauri CLI (production)
Download the installer from Releases:
.exe— NSIS installer (recommended).msi— Windows Installer
Other platforms: The source code builds on Linux and macOS via
npm run tauri build, but only Windows is officially tested and supported.
Plug & Play: Just install and launch. The setup wizard auto-detects all 12 supported local backends (Ollama, LM Studio, vLLM, KoboldCpp, llama.cpp, LocalAI, Jan, GPT4All, text-generation-webui, TabbyAPI, Aphrodite, SGLang). Nothing installed yet? The wizard shows one-click install links for every backend.
New to Locally Uncensored? Read the Getting Started Guide with screenshots for every step.
git clone https://github.com/PurpleDoubleD/locally-uncensored.git
cd locally-uncensored
npm install
npm run devgit clone https://github.com/PurpleDoubleD/locally-uncensored.git
cd locally-uncensored
setup.batInstalls Node.js, Ollama, downloads an uncensored model, launches the app.
Open the Create tab. ComfyUI is auto-detected or one-click installed. Models download with one click. Workflow is set to Auto — just write a prompt and hit Generate.
| Model | VRAM | Best For |
|---|---|---|
| GLM-4.7-Flash IQ2 | 12 GB | Strongest 30B class. Tool calling. 198K context. |
| Gemma 4 E4B | 4 GB | Lightweight, fast, great for small GPUs. |
| Qwen 3.5 9B | 8 GB | Strongest reasoning + coding at 9B. |
| Gemma 4 31B | 16 GB | Frontier dense model, native tools + vision. |
| Qwen 3.5 35B MoE | 16 GB | Best agentic, 256K context. SWE-bench leader. |
| Hermes 3 8B | 6 GB | Agent Mode. Uncensored + tool calling. |
| DeepSeek R1 (8B-70B) | 6-48 GB | Chain-of-thought reasoning. |
| Model | VRAM | Notes |
|---|---|---|
| FLUX.1 Schnell / Dev | 8-10 GB | Best text-to-image. Fast (schnell) or quality (dev). |
| FLUX 2 Klein 4B | 8-10 GB | Next-gen, fastest FLUX model. |
| Z-Image Turbo | 10-16 GB | Uncensored, 8-15 sec per image. |
| Juggernaut XL V9 | 6 GB | Best photorealistic SDXL. |
| Model | VRAM | Notes |
|---|---|---|
| Wan 2.1 T2V 1.3B | 8-10 GB | Fast entry point, 480p. |
| Wan 2.1 T2V 14B | 12+ GB | High quality, 720p. |
| FramePack F1 (I2V) | 6 GB | Image-to-video, revolutionary low VRAM. |
| AnimateDiff Lightning | 6-8 GB | Ultra-fast 4-step animation. |
| HunyuanVideo 1.5 | 12+ GB | Excellent temporal consistency. |
- Plug & Play Setup (auto-detect 12 local backends, one-click install links)
- Codex Coding Agent
- MCP Tool Registry (13 tools)
- Granular Permissions (7 categories)
- File Upload + Vision
- Thinking Mode (provider-agnostic)
- Model Load/Unload from header
- Multi-Provider (20+ presets)
- Agent Mode + Workflows
- Memory System
- A/B Compare + Local Benchmark
- RAG / Document Chat
- Voice Chat (STT + TTS)
- ComfyUI Plug & Play (auto-detect, one-click install)
- 20 Image + Video Model Bundles
- Image-to-Image (I2I)
- Image-to-Video (I2V) — FramePack, CogVideoX, SVD
- Z-Image + FLUX 2 support
- Dynamic Workflow Builder (14 strategies)
- VRAM-Aware Model Filtering
- Think Mode in Chat Input
- Voice Mode (Qwen Omni live voice)
- Upscale + Inpainting
- Plugin System
git clone https://github.com/PurpleDoubleD/locally-uncensored.git
cd locally-uncensored
npm install
npm run dev # Development
npm run tauri build # Production binary| Platform | Status | Download |
|---|---|---|
| Windows (10/11) | Fully tested | .exe / .msi |
| Linux / macOS | Build from source | npm run tauri build |
Check out the Contributing Guide. See open issues or the Roadmap.
AGPL-3.0 License — see LICENSE.
Your data stays on your machine.