Guaardvark

Guaardvark is a local AI studio for people who want document chat, task automation and media production on their own hardware. It brings those jobs into one interface and coordinates their use of the GPU.

• Chat with your files. Ask questions across project documents using keyword and vector search, reranking, and citations to source files, sections and PDF pages.
• Desktop and coding agents. Agents can operate an Ubuntu desktop and browser, while parallel coding agents work in separate Git worktrees through Ollama or Claude Code.
• Images and video. Generate with models including FLUX.1, SDXL, Wan 2.2 and CogVideoX, edit images, train character LoRAs, and carry character identities across stills and video.
• Music and voice. Create songs with ACE-Step, generate speech and sound effects, transcribe with Whisper.cpp, and clone voices with explicit consent. Music-video tools match cuts to beats, and a built-in editor handles the resulting footage.
• Editor connections. The MCP server lets tools such as Cursor, Claude Code, Codex and Zed use the studio's generation and retrieval capabilities; its MCP client lets chat use tools from other servers.

Guaardvark is free, open source under the MIT license, and runs on Linux with your own GPU through Ollama, ComfyUI and its GPU services. It works offline once models are downloaded, needs no cloud account for local operation, and sends no telemetry. Web access and posting require you to enable or initiate outbound activity; choosing Claude Code for coding agents uses that external service. Video hardware needs vary by model, with the listed Wan and CogVideoX options requiring roughly 11 to 16 GB of VRAM. Scanned documents need OCR, which isn't included by default.