[FoxlightAI™](/) [Vision](/vision)[Products](/products)[Open](/open)[Docs](https://docs.foxlight.ai/)[Developers](https://developers.foxlight.ai/)[Blog](/blog) Early Access [FoxlightAI™](/) [Vision](/vision)[Products](/products)[Open](/open)[Docs](https://docs.foxlight.ai/)[Developers](https://developers.foxlight.ai/)[Blog](/blog)[Sovereign](https://sovereign.foxlight.ai/) Foxlight™ SKULK™Open Source # One AI computer, woven from the hardware you _already have._ Skulk™ is downloadable Foxlight™ software for language, speech, vision, image, video, and music generation. Connect your Apple Silicon, AMD, and NVIDIA machines into one compute fabric, with pooled memory, compatible APIs, and models placed on hardware that can run them. Developed and published by FoxlightAI Inc. [Get Skulk™](#install)[Read the docs](https://foxlight-foundation.github.io/Skulk/) ![Skulk™ cluster dashboard with Apple, AMD, and NVIDIA nodes](/skulk-cluster-dark.webp) The Skulk™ cluster view: node health, memory, and compute across your machines. INSTALL SKULK™ ## From download to a ready node. The packaged apps include the matching runtime and dashboard. Start with one machine, confirm it works, then install the same release on each additional node. **Current stable release**Live version details are temporarily unavailable; the install channels still track stable. ### macOS app Apple Silicon · macOS 15+ Download the signed and notarized DMG, or use Homebrew. The menu-bar app gives Skulk™ one trusted identity - no separate uv, Python, or Node approval prompts. brew install --cask Foxlight-Foundation/skulk/skulk 1. Open Skulk™ from Applications. 2. Click the fox in the menu bar and choose Start Skulk™. 3. Allow Local Network access, wait for Ready, then choose Open Dashboard. [Open install guide](https://foxlight-foundation.github.io/Skulk/install/) ### Ubuntu & Debian amd64 / arm64 · desktop or headless APT installs the desktop controller and its exact matching runtime. The Foxlight™ keyring contains only the public repository key. curl -fLO https://apt.foxlight.ai/foxlight-archive-keyring.deb sudo apt install ./foxlight-archive-keyring.deb sudo apt update sudo apt install skulk 1. Type your login password when sudo asks; the terminal shows no characters. 2. Open Skulk™, choose Start Skulk™, wait for Ready, then open the dashboard. 3. For a headless node, install skulk-runtime and enable its user service. [Linux install guide](https://foxlight-foundation.github.io/Skulk/install/) ### Alternative: install from source Choose this path if you are contributing, need a source checkout, or use a Linux distribution the packages do not cover. The installer fetches the toolchain, clones stable Skulk™ into \~/skulk, builds the dashboard, and checks the node before you start it. curl -fsSL https://raw.githubusercontent.com/Foxlight-Foundation/Skulk/main/install.sh | bash cd ~/skulk uv run skulk [Read the source installation guide →](https://foxlight-foundation.github.io/Skulk/install/) ### The landscape is broken Distributed inference tooling today is fragile. Systems are buggy and experimental. Multi-node inference is painful. Speech and vision are afterthoughts. KV cache optimization is either missing or breaks models. If you've tried to run a large model across the machines you own, you've felt this. ### Skulk™ fixes it We built Skulk™ to fix the problems we hit running our own clusters: a fabric with separate compute, control, telemetry, and data planes, so inference is the first workload it carries rather than the limit of what it can. Stability claims are earned on hardware, not asserted - releases survive kill-the-master and node-bounce batteries before they ship. ARCHITECTURE ## Built to handle real workloads 🧵 ### Every Engine, One Fabric MLX in-process on Apple Silicon; llama.cpp on AMD and NVIDIA; vLLM for high-concurrency GPU serving. Skulk™ picks the engine by hardware and workload, and ships signed engine wheels with build provenance. 🔀 ### Sharding & Placement Pipeline and tensor sharding with memory-aware admission and health-aware placement. Preview placements before you commit, or place directly - you decide where compute lives. 🔊 ### Speech, First-Class Realtime speech-to-text with server-side voice detection and barge-in, text-to-speech with voice cloning, and composable speech-to-chat-to-speech pipelines. The voice of your AI mind, running locally. 🗄️ ### Model Store & Staging Download once, stage over the LAN. Revision-pinned artifacts, automated lifecycle management, and eviction that respects heterogeneous device storage. ⚡ ### Faster Decoding Multi-token-prediction speculative decoding, measured at 1.16-2.2x - on Apple Silicon and on AMD/NVIDIA alike. TurboQuant and OptiQ cache quantization squeeze more context into less memory. 🔌 ### Speaks Your API OpenAI Chat Completions, OpenAI Responses, Claude Messages, and Ollama wire formats - point the tools you already use at your own cluster. Embeddings and vision ride the same endpoint. 🎬 ### Video Generation Turn text prompts and reference images into video on compatible NVIDIA and AMD GPU nodes. Use Video Studio to plan, review, and collect renders, or work through the Skulk™ API. 🎵 ### Music Generation Generate music from prompts and optional lyrics through the Skulk™ API, using compatible models on your hardware. Track generation progress and download the finished audio. HOW IT WORKS ## From connection to compute 01 Connect your cluster Install the signed macOS app or the Ubuntu/Debian packages on each machine. Start one node, confirm it reaches Ready, then add machines on the same version and namespace. Hardware capability is detected automatically. 02 Load a model Search the built-in registry and download once to the model store. When you run a model, Skulk™ plans a memory- and health-aware placement, stages weights over the LAN, shards the model, and brings the pipeline up. 03 Run inference Use the dashboard - topology, chat with voice, observability - or point any OpenAI, Claude, or Ollama client at your cluster's endpoint. Skulk™ routes requests and handles the complexity so you don't have to. Set up a node # Apple Silicon macOS 15+ brew install --cask Foxlight-Foundation/skulk/skulk  # Ubuntu or Debian, amd64 / arm64 curl -fLO https://apt.foxlight.ai/foxlight-archive-keyring.deb sudo apt install ./foxlight-archive-keyring.deb sudo apt update sudo apt install skulk  # Open Skulk™ → Start Skulk™ → Open Dashboard STATUS ## Honest software Skulk™ is available for macOS and Linux, and the project is moving quickly. Signed macOS and packaged Linux releases pin one reviewed runtime across every artifact. Start with a one-node cluster and expand one machine at a time: distributed inference still depends on the hardware, network, model, and placement you choose, and every node in a cluster must run the same version. Video generation runs on compatible NVIDIA and AMD GPU nodes; music generation is available through the API with compatible models. Generation support depends on the model and hardware you choose. SKULK™ OPERATORMobile App ## Your cluster, in your pocket Skulk™ Operator is the mobile app for a Skulk™ cluster you own and run - topology, models, chat, and activity from your phone. It is available now on the App Store and Google Play. Pair with a QR code; every byte between the app and your cluster is end-to-end encrypted through a blind relay that Foxlight™ cannot read. [Download on the App Store](https://apps.apple.com/us/app/skulk-operator/id6800721590)[Get it on Google Play](https://play.google.com/store/apps/details?id=ai.foxlight.skulk)[Explore Skulk™ Operator](/products/skulk-operator) [Install Skulk™](#install) FoxlightAI™ Products [Skulk™](/products/skulk)[Sovereign](https://sovereign.foxlight.ai/)[Skulk™ Operator](/products/skulk-operator)[Foxmemory™](/products/foxmemory)[Foxden™](/products/foxden)[Open Source](/open) Company [Vision](/vision)[Foundation](/foundation)[Docs](https://docs.foxlight.ai/)[Developers](https://developers.foxlight.ai/)[Skulk™ Operator Support](/support/skulk-operator)[Call us: 1 903-776-4387 · Voice only](tel:+19037764387) Legal [Privacy](/legal/privacy)[Terms](/legal/terms)[GitHub](https://github.com/Foxlight-Foundation) © 2026 FoxlightAI Inc. Foxlight, FoxlightCore, FoxlightAI, Skulk, the Foxlight logo, and the Skulk logo are trademarks of FoxlightAI Inc.