ArcFlow
The World Model Engine
A game engine simulates a world. ArcFlow records the real one.
ArcFlow records what's actually happening in a physical space — every entity, position, relationship, and event, each with a confidence score and full history — and lets you query any moment, past or present, in one dialect. Graph storage, query execution, live views, an event bus, a behavior engine, an algorithm library, durability, and language bindings all live inside one engine, designed from scratch for spatial-temporal workloads — not an assembled stack. One dialect describes the schema, the query, the view, the trigger, and the algorithm call.
Why one engine, not a stack
Most stacks for the physical world are assemblies. A graph database here, a streaming engine there, a message bus, a workflow runner, an algorithm library. Each speaks its own dialect. Each carries its own schema. The integration cost is paid every day, by every engineer on the team.
ArcFlow is the alternative. One engine where every layer shares the same data model and the same query dialect. You write the schema, the query, the live view, the event subscription, the behavior trigger, and the algorithm call in one language — and they reach the same graph, on the same machine, at frame rate.
That coherence is the product. The latency numbers — sub-millisecond where assembled stacks measure in tens — are the consequence.
Eight layers, one query language
World Graph
Typed entities, relationships, indexes.
Query Engine
Compile, plan, execute.
Live Surface
Standing queries that update themselves.
Event Bus
Topics, subscriptions, request-reply.
Behavior Engine
Triggers, behavior trees, durable workflows.
Algorithm Library
Twenty-nine graph, spatial, and fusion routines.
Two more layers sit underneath. The World Store is the storage substrate — write-ahead log, snapshots, manifests, and replication; explicit, visible, and free of magic. The Bindings layer exposes a C ABI with first-class wrappers for Python, Rust, TypeScript, and Go, so the same semantics reach every language.
Runs everywhere
Same platform, different scale
Native, Docker, Astro, mobile, edge, browser.
Every ArcFlow instance is a fragment that operates independently and syncs when connected. Run it on a stadium server processing 22 players at 60fps. Run it on a laptop during development. Run it on a phone for field work. The queries are identical. The platform is identical. Only the scale changes.
ArcFlow is going open-source under BSL-1.1. The platform is free. The advantage we keep is everything above it — the hardware, the perception models, the operational playbooks, the venue...