Get Started

Neural world models simulate how the world evolves under actions. ArcFlow records what actually happened. These are different tools for different jobs — simulation in the neural tier, persistence here.

ArcFlow is the operational world model layer. A persistent, spatially indexed, temporally versioned, confidence-scored store of actual world state — exact, queryable in microseconds, running in-process alongside whatever generates the data.

MATCH (s:Sensor)-[d:DETECTED]->(r:Robot)-[:OCCUPIES]->(z:Zone)
WHERE d.confidence >= 0.90 AND z.hazard > 0.15 AND r.battery < 30
RETURN r.name, z.name, avg(s.reliability * d.confidence) AS trust

This runs in ~0.1ms. Deterministic. Against exact history. ArcFlow tells you what did happen — which sensor, which detection, what confidence, which zone, at any moment in the past — with one query.

The AI infrastructure stack has three tiers. Foundation models handle language and perception. Neural world models simulate how the world evolves under actions. ArcFlow is the persistence tier — the layer where actual world state lives, versioned, spatially indexed, confidence-scored, queryable.

It runs in-process — no server, no network hop, no inference cluster. Neural world models require GPU infrastructure and return in 100–1000ms. ArcFlow returns in microseconds, co-located with the agent, robot, or vehicle that needs it. For any real-time control loop, this is an architectural requirement, not a convenience.

Every mutation is versioned via ArcFlow Clock Domains — each source advances at its own rate, the world model merges them. Every fact is graded by the ArcFlow Evidence Model — observation class (observed, inferred, predicted), confidence score, and provenance chain are storage primitives, not optional annotations. The full history is queryable with ISO GQL.

Try instantly — zero install#

Open ArcFlow Engine in your browser. Paste any GQL query — persistent storage, no install, no signup.

CREATE (e:Entity {id: 'unit-01', x: 12.4, y: 8.7, _observation_class: 'observed', _confidence: 0.94})
CREATE (e:Entity {id: 'contact-x', x: 80.0, y: 90.0, _observation_class: 'predicted', _confidence: 0.38})
CALL algo.nearestNodes(point({x: 0, y: 0}), 'Entity', 5) YIELD node, distance RETURN node.id, distance
MATCH (e:Entity) WHERE e._confidence > 0.8 RETURN e.id, e.x, e.y
MATCH (e:Entity) AS OF seq 1 RETURN e.id, e._confidence

Install#

curl -fsSL https://oz.com/install | sh

Then:

arcflow --playground

All paths#

Next#

• Quickstart — first world model query, step by step
• Installation — all install methods in detail
• Project Setup — Express, testing, monorepo

Guides#

• Building a World Model — entities in space, temporal memory, live monitoring
• Swarm & Multi-Agent — shared world model as coordination substrate for AI agent swarms
• Trusted RAG — confidence-filtered retrieval, graph context, vector + graph + full-text
• Agent Integration — CLI binary, MCP, and napi-rs — which surface fits your execution context
• Data Quality — live DQ, batch==delta equivalence, pipeline integrity
• Code Intelligence — queryable codebase graph for coding agents

Use cases#

• Autonomous Systems — robot fleets and UAV coordination on a shared world model
• Digital Twins — live spatial replica of physical facilities
• Robotics & Perception — sensor fusion, track lifecycle, ROS bridge