guaardvark: Run a Local AI Studio on Your GPU

Overview

Guaardvark is a self-hosted AI studio that brings local chat, document retrieval, coding agents, and media generation into one application. Its goal is to coordinate workloads on your own machine rather than rely on cloud services for the core experience.

It is aimed at people who want a broad set of AI workflows in one place, including through a browser interface, command-line tools, or MCP connections to coding agents. It makes the most sense when local data control and a single integrated workstation matter more than a minimal install or a hosted, low-maintenance service.

Key Features

• Generates images, video, music, and speech locally, with GPU-aware resource coordination.
• Provides hybrid document and code retrieval with citations, reranking, and project-level knowledge bases.
• Runs coding-agent swarms in isolated Git worktrees, with dependency-aware merging and optional local-model backends.
• Connects to external coding agents through an MCP server and agent skills.
• Offers screen agents that operate a virtual Ubuntu desktop and report their actions.
• Includes media workflows such as a multi-role Film Crew, beat-synced music-video generation, and a built-in video editor.
• Supports offline operation through Flight Mode and supervised controls for selected automation.

Use Cases

• A local-AI enthusiast can use one workstation for chat, document search, and media generation when keeping data on-device is important.
• A developer can connect a coding agent to repository tools or run parallel coding tasks when isolated worktrees and local control are useful.
• A video creator can generate clips, voice, or music and combine them in the studio when a capable GPU is available.
• A team or individual with a large document collection can build a searchable knowledge base with source citations for research and project work.

Getting Started

On Linux, install and launch with:

curl -fsSL https://guaardvark.com/install.sh | bash

See the README and installation guide for manual setup, platform notes, and configuration.

Alternatives

• MaxKB: MaxKB focuses on knowledge-grounded agents, workflows, and business deployments, while Guaardvark also combines media generation, coding agents, and automation.

Considerations

• The documented primary target is Linux with an NVIDIA GPU. The README lists 8 GB or more of VRAM as a requirement and recommends 16 GB for video generation.
• Python 3.12 is required. Setup also includes services such as PostgreSQL, Redis, and Ollama, so this is a substantial workstation install rather than a lightweight standalone app.
• Apple Silicon support is partial, with GPU features arriving through Metal; Windows support through WSL2 is being verified.
• Some capabilities depend on separate models, plugins, or system tools. For example, video editing requires MLT/Shotcut components, and some model downloads have additional licensing conditions.
• The project has 244 stars and 11 open issues in the supplied repository data. Its broad feature set and hardware dependencies are worth evaluating against the workflows you actually need.