An order of computer witches specializing in cloud native technologies.

We build infrastructure for private AI deployment on customer-owned hardware. Our work sits at the intersection of distributed systems, GPU inference, and quantization research.

Geodesic is a distributed LLM inference orchestrator written in Rust. It schedules and serves open-weight models across heterogeneous GPU clusters using gossip-based coordination and MIP-optimized placement. Built on mistral.rs.

Private AI Cells are complete, air-gappable AI stacks for organizations that need to run frontier models on hardware they control. Targeting regulated verticals where data sovereignty is non-negotiable: legal, healthcare, government.

• Distributed GGUF inference across mixed GPU and Apple Silicon fleets
• Domain-calibrated quantization that preserves safety alignment
• Gossip-based cluster coordination (libp2p, Kademlia DHT)
• MoE expert caching, offloading, and routing-aware scheduling
• Air-gap deployment and enterprise packaging

Nova Scotia, Canada