One Binary. Every Platform.

• Windows 10 / 11 (x86_64)
• Windows Server 2019+
• Native MSVC build
• ProjFS virtual filesystem integration
• WinFsp mount support
• DirectX 12 / Vulkan GPU acceleration

• x86_64 (Intel/AMD)
• aarch64 / ARM64 (Raspberry Pi 4+, Graviton)
• Fedora, Ubuntu, Debian, EndeavourOS, Alpine
• FUSE virtual filesystem
• CUDA / Vulkan GPU acceleration
• Systemd service integration

• Apple Silicon (M1/M2/M3/M4)
• macOS 13 Ventura+
• Native ARM64 build
• macFUSE virtual filesystem
• Metal GPU acceleration
• Apple Neural Engine (ANE) support

• iOS 16+ (ARM64)
• Android 10+ (ARM64, x86_64)
• Embedded library mode
• Battery-aware resource governor
• Background sync via OS-native APIs

• Raspberry Pi 3/4/5 (ARM64)
• ESP32-S3 (flash binary)
• NVIDIA Jetson (ARM64 + CUDA)
• Intel NUC / Mini PCs
• Any Linux with 256MB+ RAM

• Native OS where the kernel IS the operating system
• Direct hardware access — no host OS overhead
• Own filesystem, drivers, and HAL
• Coupling-aware process scheduling
• Currently in research & design

NVIDIA GPU acceleration for coupling matrix evaluation, vector distance computation, and batch processing. Requires CUDA toolkit. Supports RTX 20/30/40/50 series, A100, H100, and all compute-capable NVIDIA GPUs.

Cross-platform GPU compute via Vulkan. Works on NVIDIA, AMD, and Intel GPUs. WGSL compute shaders for coupling evaluation, normalization, and distance computation. No vendor lock-in.

Apple Silicon GPU acceleration via Metal. Optimized for M-series unified memory architecture. Shared memory between CPU and GPU eliminates data transfer overhead for coupling evaluation.

Neural processing unit acceleration for inference workloads. Supports Apple Neural Engine (ANE), Intel NPU, Qualcomm Hexagon DSP, and Google Edge TPU. Power-efficient AI inference for edge devices.

Platform-native virtual filesystem integration. ProjFS on Windows, FUSE on Linux, macFUSE on macOS. Database entities appear as files in your filesystem — open, read, and write with any tool.

Optional protocol modules for gRPC (Protocol Buffers) and GraphQL query interface. Loaded as modules — not compiled into the kernel. Available when needed, zero overhead when not.

Peer-to-peer federation module for decentralized mesh networking. NAT traversal, hole punching, and direct peer connections. Loaded when cluster mode is enabled. Not needed for single-node deployments.

Quantum computing integration for optimization verification. Connects to IBM Quantum, Origin Quantum Cloud, and D-Wave for cross-platform quantum validation of coupling matrix ground states. Module loaded on demand.

1. Place the binary anywhere on your system
2. Run it — the first-boot wizard opens in your browser
3. Configure storage path, network settings, and identity
4. The kernel auto-detects hardware and pulls acceleration modules
5. Start ingesting data immediately

No package manager, no dependency chain, no apt/brew/choco needed. Just the binary.

1. Flash the binary directly to the device's storage
2. Configure via node.toml (text file, minimal settings)
3. Device boots directly into the coupling engine
4. Connects to seed nodes for network discovery
5. Pulls modules and starts contributing to the network

For IoT gateways, edge devices, and embedded deployments. Headless mode with CLI management.

New versions propagate through the network's gossip protocol. Every node receives updates automatically. Post-quantum signature verification (ML-DSA-65) ensures only authentic binaries are installed. Atomic swap prevents corruption.

Acceleration modules update independently of the kernel. A new GPU module deploys without restarting the engine. Module manifests track compatibility — the kernel only loads modules that match its version.

If a new version causes issues, the kernel automatically rolls back to the previous version. The .bak binary is preserved during every update. Rollback is instant and lossless.

pip install vectorscaledb

Available at launch

cargo add vectorscale-client

Available at launch

npm install @vectorscaledb/client

Available at launch

The MCP server exposes every VectorScaleDB capability as a tool that AI agents can discover and use. Ingest, query, analyze, explore — all available via the stdio transport protocol. Works with Claude Code, Claude Desktop, and any MCP-compatible agent.

• Client: Proxy to a running VectorScaleDB instance
• Standalone: Embedded engine with local storage
• Network: Embedded engine with federation