VisionFlow

Real-Time 3D Knowledge Graphs with Autonomous AI Agents

100,000+ nodes at 60 FPS | 55x GPU acceleration | 80% bandwidth reduction

Quick Start | Documentation | Benchmarks | Roadmap

---

What is VisionFlow?

VisionFlow transforms static documents into living knowledge ecosystems. Deploy autonomous AI agents that continuously analyse your data, discovering connections while you explore results in an immersive 3D space with your team.

Your Data --> AI Agent Teams --> Knowledge Graph --> 3D Visualisation --> Team Collaboration

Why Choose VisionFlow?

Feature	VisionFlow	Traditional Tools
Performance	60 FPS @ 100K nodes (GPU)	4-15 FPS (CPU only)
Latency	10ms WebSocket updates	50-100ms+
Privacy	Self-hosted, your infrastructure	Third-party APIs
Collaboration	Real-time multi-user 3D	Single-user text output
Intelligence	Autonomous 24/7 AI agents	Reactive query-based
Auditability	Git version control	No transparency

---

Quick Start

Deploy in under 60 seconds:

git clone https://github.com/DreamLab-AI/VisionFlow.git
cd VisionFlow && cp .env.example .env
docker-compose --profile dev up -d

Access your instance:

Service	URL	Description
Frontend	http://localhost:3001	3D visualisation interface
Neo4j Browser	http://localhost:7474	Graph database explorer
API	http://localhost:4000/api	REST & WebSocket endpoints

Your AI agents begin analysing data immediately.

Native Installation (Rust + CUDA)

# Prerequisites
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
# Install CUDA 12.4: https://developer.nvidia.com/cuda-downloads

# Build and run
git clone https://github.com/DreamLab-AI/VisionFlow.git
cd VisionFlow && cp .env.example .env
cargo build --release --features gpu
cd client && npm install && npm run build && cd ..
./target/release/webxr

---

Key Features

GPU-Accelerated Physics 100+ CUDA kernels across 13 modules 55x faster than CPU physics 100K+ nodes at 60 FPS Force-directed layouts in real-time	Autonomous AI Agents 50+ concurrent agents with specialised roles Microsoft GraphRAG integration Multi-hop reasoning across domains Continuous 24/7 knowledge discovery	Ontology Intelligence OWL 2 EL reasoning via Whelk-rs 10-100x faster inference Semantic physics for meaningful layouts Automatic contradiction detection
Binary Protocol V2 36-byte messages (80% smaller than JSON) 6.9x lower latency end-to-end Sub-10ms WebSocket updates Zero-copy serialisation	Real-Time Collaboration Multi-user synchronisation Independent camera controls Voice-to-voice AI interaction Spatial audio via WebRTC	Immersive XR Meta Quest 3 support (Beta) Hand tracking & controllers Force-directed 3D physics Vircadia integration (planned)

---

Performance

Benchmarks (AMD Ryzen 9 5950X + RTX 4080)

Metric	Result	Notes
Max Nodes @ 60 FPS	180,000	GPU-accelerated physics
Max Nodes @ 30 FPS	450,000	Instanced rendering
Physics Speedup	55x	GPU vs multi-threaded CPU
WebSocket Latency	10ms	Binary protocol V2
Bandwidth Reduction	80%	3.6 MB vs 18 MB per frame
Concurrent Users	250+	Stress-tested 48 hours

GPU vs CPU Physics (100K nodes, 200K edges)

Operation	GPU (CUDA)	CPU	Speedup
Force Calculation	2.3ms	145ms	63x
Position Update	0.4ms	12ms	30x
Collision Detection	1.8ms	89ms	49x
Total Frame	4.5ms	246ms	55x

View Detailed Benchmarks

Scalability by Graph Size

Nodes	Edges	Frame Time	FPS	GPU Memory
1K	2K	0.08ms	12,500	4 MB
10K	20K	0.5ms	2,000	40 MB
100K	200K	4.5ms	222	400 MB
500K	1M	18ms	56	2 GB
1M	2M	35ms	29	4 GB

Comparison with Alternatives (100K nodes)

Solution	FPS	Latency	Bandwidth	GPU Memory
VisionFlow	60	10ms	3.6 MB	400 MB
Neo4j Bloom	25	45ms	18 MB	1.2 GB
GraphXR	35	28ms	8 MB	650 MB
Gephi	8	N/A	N/A	N/A
Cytoscape	12	N/A	N/A	N/A

Full benchmarks: docs/reference/performance-benchmarks.md

---

Architecture

flowchart TB
    subgraph Client["Client Layer"]
        React["React + Three.js"]
        WebXR["WebXR / Quest 3"]
        Voice["Voice UI"]
    end

    subgraph Server["Rust Server (Actix)"]
        subgraph Actors["Actor System"]
            GS["GraphState"]
            PO["PhysicsOrchestrator"]
            SP["SemanticProcessor"]
            CC["ClientCoordinator"]
        end
        HP["Hexagonal Ports"]
    end

    subgraph Data["Data Layer"]
        Neo4j[(Neo4j 5.13)]
    end

    subgraph GPU["GPU Compute"]
        CUDA["100+ CUDA Kernels"]
    end

    Client <-->|"36-byte Binary Protocol"| Server
    Server <--> Neo4j
    Server <--> CUDA

    style Client fill:#e1f5ff,stroke:#0288d1
    style Server fill:#fff3e0,stroke:#ff9800
    style Data fill:#f3e5f5,stroke:#9c27b0
    style GPU fill:#e8f5e9,stroke:#4caf50

Loading

Core Design Principles

Principle	Implementation
Server-Authoritative	Neo4j is the single source of truth
Binary Protocol	36-byte WebSocket messages (80% bandwidth reduction)
GPU Offloading	Physics, clustering, pathfinding accelerated 55x
Actor System	21 specialised actors for state, physics, semantics
Event-Driven	Domain events with pub/sub for loose coupling

Deep Dive: Architecture Overview | Hexagonal CQRS | Actor System

---

Technology Stack

Layer	Technology
Backend	Rust 1.75+, Actix-web 4.11, Hexagonal Architecture
Database	Neo4j 5.13 (neo4rs)
GPU Compute	CUDA 12.4, cudarc, cust (100+ kernels)
Reasoning	OWL 2 EL, Whelk-rs, horned-owl
Frontend	React 18, Three.js (React Three Fiber), TypeScript
XR	Babylon.js, WebXR, Meta Quest 3
AI	MCP Protocol, Claude, Microsoft GraphRAG
Networking	Binary WebSocket, QUIC/WebTransport (quinn)

---

Use Cases

Research Teams

• Literature review automation - AI agents continuously analyse papers, surface connections
• Ontology development - OWL reasoning validates knowledge structures
• Collaborative exploration - Multi-user 3D navigation of research landscapes

Enterprise Knowledge Management

• Corporate knowledge graphs - Integrate documents, wikis, databases
• AI-assisted discovery - Find hidden relationships across silos
• Audit compliance - Git-versioned changes with full provenance

AI/ML Pipelines

• GraphRAG integration - Enhance LLM context with structured knowledge
• Training data curation - Visual exploration of dataset relationships
• Model interpretability - Visualise attention patterns and embeddings

Healthcare & Life Sciences

• SNOMED CT / Gene Ontology - Whelk-rs handles 354K+ class ontologies
• Clinical decision support - Real-time reasoning over medical knowledge
• Drug discovery - Visualise compound-target-pathway relationships

---

Documentation

VisionFlow uses the Diataxis framework for organised documentation:

Type	Purpose	Start Here
Tutorials	Learning-oriented	Installation, First Graph
How-To Guides	Task-oriented	Agent Orchestration, XR Setup
Explanations	Understanding-oriented	Architecture, Semantic Physics
Reference	Information-oriented	REST API, WebSocket Protocol

Complete Documentation: docs/

---

Roadmap

Current (v2.0)

• Modular actor architecture (21 actors)
• Neo4j as primary database
• Binary WebSocket protocol V2 (36 bytes)
• 100+ CUDA kernels (55x speedup)
• OWL 2 EL reasoning with Whelk-rs
• Meta Quest 3 WebXR (Beta)

In Progress (v2.1)

• Vircadia multi-user integration
• SPARQL query interface
• Distributed GPU compute
• Nostr identity integration

Future (v3.0+)

• Apple Vision Pro native app
• Federated ontologies
• Kubernetes operator
• W3C DID for agents

Detailed Roadmap: See full roadmap with quarterly sprints

---

Contributing

We welcome contributions! See our Contributing Guide.

# Fork and clone
git clone https://github.com/YOUR_USERNAME/VisionFlow.git
cd VisionFlow

# Setup
cargo build && cd client && npm install && cd ..

# Test
cargo test && npm test

# Submit PR

Contribution Areas: Bug Fixes | Documentation | Features | Performance | Testing

---

System Requirements

Tier	CPU	RAM	GPU	Use Case
Minimum	4-core 2.5GHz	8 GB	Integrated	Development, < 10K nodes
Recommended	8-core 3.0GHz	16 GB	GTX 1060 / RX 580	Production, < 50K nodes
Enterprise	16+ cores	32 GB+	RTX 4080+ (16GB VRAM)	100K+ nodes, multi-user

Platform Support: Linux (full), macOS (CPU-only), Windows (WSL2), Meta Quest 3 (Beta)

---

Acknowledgements

Built on the work of:

• 3d-force-graph - Force-directed visualisation
• graph_RAG - Natural language queries
• JavaScriptSolidServer - Solid Protocol (AGPL-3.0)

Special thanks to Prof. Rob Aspin for research in immersive knowledge visualisation.

---

License

Mozilla Public License 2.0 - Use commercially, modify freely, share changes to MPL files.

Full License

---

Transform how you discover knowledge.

git clone https://github.com/DreamLab-AI/VisionFlow.git && cd VisionFlow && docker-compose --profile dev up -d

Documentation | Issues | Discussions

Built by the VisionFlow Team