mvec is an standalone or app-embeddable vector database for Apple Silicon, designed for developers and power users to tap into a device's full GPU capabilities.
A three-model neural architecture. Each trained on different datasets of a common problem space. Disagreements represent new information with lower correction-cost-to-improvement than traditional approaches.
An execution engine for Intelligence pipelines as directed graphs. Search, infer, reason, train, coordinate — executed in parallel across CPU and GPU cores.
A technical paper exploring a radix hash index that serves CPU directed probes and GPU batch operations from the same unified-memory layout.
Unified memory architecture — the strategic case for building a CPU/GPU database from the ground up for Apple Silicon before they move into the datacentre game.
Vectors, streams, trees, graphs, tensors, documents. Redis speed, git style commits, thread-safe, zero-copy, distributed across networks using wire-optimised deltas.
A milestone-driven implementation and benchmarking plan for a radix hash index across SIMD grouping, overflow strategies, and probabilistic membership.
Graph neural networks as the execution layer for intelligent search — a dual-agent split between fast list manipulation and deep knowledge construction.
Personalisation without loss of control, collaboration without central trust, information integrity in a world of misinformation and disinformation.
Shipping Rust, Protobuf, Go, TypeScript, JSON config, SQL migration, Python validation, Dockerfile, Terraform, and Bash.
The msearch/mgraph stack a distributed database for complex datatypes, zero copy from wire to compute (GPU on Apple Silicon). This is our benchmark plan.
We reply within two business days. If a call would be faster, book a thirty minute conversation.