Lucebox
Lucebox provides a dedicated, headless AI workstation that runs open-weights models entirely on-premises, keeping proprietary code and data within the user's network. The system pairs a Ryzen AI MAX+ PRO 495 processor with a Radeon AI PRO R9700 GPU, using up to 192 GB of unified memory to serve large models like Qwen and DeepSeek at high speed. It exposes an OpenAI- and Anthropic-compatible API, so existing tools such as Claude Code, Codex, and OpenCode can connect without modification.
- Artificial Intelligence
- Developer Tools
- Hardware
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Organizations and developers increasingly rely on large language models for coding and knowledge work, but cloud-based AI services require sending proprietary code, documents, and prompts to third-party data centers. This creates privacy risks, ongoing usage costs, and dependence on external providers that can change pricing, retire models, or tighten limits at any time.
Solution
Lucebox provides a dedicated, headless inference workstation that runs open-weights models entirely on-premises, keeping all data within the user's network. The hardware combines a Ryzen AI MAX+ PRO 495 processor with a Radeon AI PRO R9700 GPU, using up to 192 GB of unified memory to serve large models like Qwen and DeepSeek at high speed. The system exposes an OpenAI- and Anthropic-compatible API, so existing tools such as Claude Code, Codex, and OpenCode can point to it without modification. Setup is handled through a browser-based flow over encrypted Bluetooth, and ongoing management occurs through a web interface served by the workstation itself. The inference engine is open source under Apache 2.0, and the system is optimized through hardware-software co-design to achieve fast prefill and decode speeds.
Target Audience
Primary customers are software development teams, AI engineers, and organizations that need private, high-performance local inference for coding agents and LLM workloads without sending proprietary data to cloud providers.
Features
- Dual-processor architecture with Ryzen AI MAX+ PRO 495 (up to 192 GB unified memory at 273 GB/s) and Radeon AI PRO R9700 (32 GB GDDR6 at 640 GB/s) working in parallel on every token
- Speculative decoding with DFlash2 block-diffusion drafter achieving up to 208 tok/s decode on Qwen3.8-27B, 3.8x faster than llama.cpp with the same drafter
- OpenAI- and Anthropic-compatible API endpoints that work with Claude Code, Codex, OpenCode, and other standard clients
- Headless Ubuntu Server with browser-based setup via encrypted Bluetooth (AES-256-GCM) and ongoing management through Lucebox Manage web interface
- Optional Tailscale integration for secure remote access over WireGuard tunnel with zero open ports
- 4 TB NVMe storage (up to 7,400 MB/s read), 10 GbE connectivity, and optional 100 GbE cluster kit
- Open source inference engine (Apache 2.0) with 60+ contributors and models available on Hugging Face