Self-host Guaardvark
v3.4.2Self-hosted AI studio with agents, video, image, music
- Stars
- 292
- CPU
- 2 CPU cores (minimum), 8 CPU cores (recommended)
- RAM
- 4 GB RAM (minimum), 16 GB RAM (recommended), 32 GB+ RAM (optimal)
- Storage
- 10 GB disk (minimum), 50 GB disk (recommended), NVMe SSD (optimal)
- Ports
- 5173 (Web UI (Vite dev server)), 5000 (API (Flask backend))
- Requires
- Node.js, Ollama, PostgreSQL, Python, Redis
What is Guaardvark?
Guaardvark is a self-hosted AI platform that turns a Linux machine with an NVIDIA GPU into a complete local AI studio. It provides a web interface and REST API for chatting with your own documents using hybrid keyword/vector search with citations, running autonomous agents that control a virtual Ubuntu desktop to browse, click and type, managing swarms of parallel coding agents in separate git worktrees, and generating video, images, music, and voice — all on your own hardware. A built-in MCP server exposes over 50 tools to popular code editors, while an MCP client lets the chat call external MCP servers. The stack includes PostgreSQL with pgvector, Redis, Ollama, ComfyUI, and a plugin system for GPU services. Installation is a single curl command that sets up Python 3.12, Node 22, PostgreSQL, Redis, Ollama, Whisper.cpp and the frontend. macOS Apple Silicon and Windows WSL2 are supported with some GPU limitations. The project is MIT licensed, developed by Albenze, Inc., with no telemetry and all outbound network paths declared in egress.json and off by default.
Should you self-host Guaardvark?
Guaardvark scores 53/100 on selfhost.directory (how scoring works) and ranks #81 of 209 in AI Agents. It has 292 GitHub stars.
It shipped 10 stable releases in the past year. The latest, v3.4.2, came out on 27 October 2026.
Setup is advanced: plan for several services and some ongoing maintenance. It is MIT-licensed and written mostly in Python.
Its closest alternative on selfhost.directory is AnythingLLM (88/100, 67 669 stars). It scores higher overall, so compare the two before you pick. See Guaardvark vs AnythingLLM side by side.
Guaardvark README
FAQ
Is it free?
Yes. Guaardvark is open source under the MIT License, and running it costs nothing beyond your own hardware and power.
How is it different from ComfyUI or Open WebUI?
They are excellent at one slice each. Guaardvark puts chat, RAG, agents, coding swarms and the whole media pipeline in one install that shares one GPU, and it hands off to ComfyUI with one click when you want the node graph.
Can I use it from Claude Code, Cursor or Codex?
Yes. Install the Claude Code plugin with the two lines in Quick start, or run `python -m backend.mcp install` to wire the MCP server into every agent client it finds.
Does it run on macOS or Windows?
Apple Silicon is supported, with GPU features arriving through Metal; see INSTALL.md. Windows through WSL2 is being verified.
What hardware do I need?
Linux with one NVIDIA GPU. Chat, RAG, voice and the coding swarm run CPU-only; 8–12 GB adds vector RAG, screen agents, images, upscaling and music; video generation needs a 16 GB-class card.
Does it work offline?
Yes, once the models you use are downloaded. Only features that reach other services by design, such as web search, posting and the Discord bot, need a connection.
Does Guaardvark send my data anywhere?
No, not unless you turn something on. Models run locally, and every outbound path (model downloads, web search, posting, the Discord bot, a remote Ollama) is declared in egress.json and stays off until a setting or a person starts it.
More in Image Generation
InvokeAISelf-hosted, open-source AI image generation studio with a layered Unified Canvas, node-based workflows, ControlNet, and broad local model support from SD 1.5 to Flux.
JaazAn open-source AI creative agent with an infinite canvas for one-prompt image and video generation, a Canva alternative that runs locally on Windows and macOS.
LanceOpen-source, self-hosted unified multimodal model from ByteDance: 3B active parameters covering text-to-image, video synthesis, media editing, and visual Q&A.
Mix StudioSelf-hosted, open-source AI workspace built on ComfyUI that runs curated image and video workflows locally with a touch-friendly web interface.
MxPageSelf-hosted AI workspace for e-commerce product images and Xiaohongshu posts. Runs locally on Windows/macOS via Electron or from source with your own AI API keys.
oxigenOxigen is a self-hosted, open-source social media image generator creating Open Graph images via web UI and REST API, as Docker container or standalone binary.
ProductFlowProductFlow is an open-source, self-hosted product creative workbench for solo merchants to manage assets, AI copy, and image workflows.
sdnextSelf-hosted, open-source Stable Diffusion web UI to generate, caption, upscale and edit images and video, with many GPU backends and model quantization for low VRAM.
stable-diffusion-webuiA self-hosted, open-source web UI for Stable Diffusion with txt2img, img2img, inpainting, outpainting, upscaling, and extensive customization options.
Stable Diffusion WebUI DockerSelf-hosted Docker container bundling 14 Stable Diffusion WebUIs like Automatic1111, ComfyUI, Fooocus, and InvokeAI, each on its own port via docker-compose profiles.
Stable Diffusion WebUI ForgeSelf-hosted, open-source Stable Diffusion web UI that improves GPU memory management and speeds up inference, with ControlNet, IP-Adapter and Flux quant support.
Sygil-webuiSelf-hosted open-source web UI for Stable Diffusion with dual Streamlit/Gradio interfaces, dynamic previews, VRAM optimization, upscalers, and text-to-video.
0 comments