Project · mgraph
mgraph
Collaboration for distributed intelligence systems: an engine for atomic coordination of complex data types — from threads to global distribution.
An embeddable engine for transactional operations on graphs, vectors, tensors, streams, and documents. Thread-safe locally. Delta-propagated across networks. The same engine, the same typed operations, from in-process to worldwide.
Collaboration solved as a primitive — any data structure, across any boundary.
The data structures that AI systems, collaborative tools, and modern applications depend on — graphs, tensors, vectors, streams — have inherent complexity that adds cost at every layer.
It is first an embeddable engine with git-like transactions and thread-safe atomic operations. Wrap it in a server and you have Redis-speed structured data over the network. Extend to a relay mesh and you have globally distributed coordination.
The same engine at every scale.
The architecture in one read.
Typed by design
The engine knows the shape of the data it carries. Graphs, vectors, tensors, streams, documents — stored and operated on natively. That awareness is what makes deltas, parallelism, and zero-copy access tractable in the same system.
Atomic operations
Git-like version model. Branch for isolation, commit atomically, merge or fast-forward. Thread-safe concurrent reads and writes on structured data without flattening it.
Delta wire format
Every mutation encodes as a semantic delta — "edge A→B added" not "bytes changed." Fixed-width header for routing, type-aware payload. Relays forward without deserialising. Cost scales with the change, not the data.
Many environments
Browser, mobile, edge, datacenter — one runtime, one contract. State synchronises wherever it needs to live.
QUIC-native transport
Per-document streams with independent ordering and backpressure. Multiplexed, congestion-aware, built for mobile and edge environments.
Where it goes from here.
- Delta engine — graphs, vectors, streamsdevelopedPer-type encoders and a unified delta container.
- QUIC transport with multiplexed channelsdevelopedProduction-ready transport for mobile and edge.
- Browser runtimedevelopedWebAssembly + service-worker mode.
- Apple Silicon runtimein flightNative engine sharing memory with msearch on the same device.
- Public docs and the open core releasein flightSource, spec, and examples; commercial layers around the open substrate.
- Tensor and document delta encodersnextRound out the type set so any structured payload can ride mgraph.
- Conflict-free coordination across regionsnextCRDT-shaped semantics on top of the delta engine.
- Sponsored partnership programmenextOpen-core partnerships with co-built commercial layers.
Tell us what you're working on.
We reply within two business days. If a call would be faster, book a thirty minute conversation.