| title | VisionClaw Documentation |
|---|---|
| description | Documentation hub for VisionClaw — a governed agentic mesh for real-time 3D knowledge-graph exploration with GPU-accelerated physics, OWL 2 EL ontology reasoning, and a multi-agent control surface. |
VisionClaw · Documentation Hub
VisionClaw is a governed agentic mesh for real-time 3D knowledge-graph exploration: a Rust backend, CUDA GPU physics, OWL 2 EL ontology reasoning over an embedded Oxigraph triple store, and a multi-agent AI control surface.
git clone https://github.com/DreamLab-AI/VisionClaw.git
cd VisionClaw && cp .env.example .env
./scripts/launch.sh up dev./scripts/launch.sh up dev is the canonical launcher. The explicit fallback is docker compose -f docker-compose.unified.yml --profile dev up -d — docker-compose.unified.yml is the only compose file shipped.
| Service | URL | Notes |
|---|---|---|
| 3D graph frontend | http://localhost:3001 | nginx |
| REST API | http://localhost:4000/api | HTTP + WebSocket |
| Solid pod | http://localhost:8484 | per-user pod storage |
| Legacy MCP (TCP) | localhost:9500 |
agent orchestration channel |
The graph store is the embedded Oxigraph triple store backed by SQLite (ADR-11). Neo4j is fully removed (ADR-132) and there is no separate database browser UI.
| Layer | Facts |
|---|---|
| Backend | 428 Rust files (~178K LOC); 35 Actix actors (19 service + 16 GPU; +10 WebSocket session); 44 hexser handlers (19 directive + 25 query), no CQRS bus (ADR-089); 9 ports, 12 adapters; 8 workspace crates |
| GPU physics | 82 CUDA __global__ kernels across 9 .cu files (5,854 LOC); 55× speedup — 246 ms CPU (4 FPS) → 4.5 ms GPU (222 FPS) at 100K nodes |
| Client | 465 TypeScript/TSX files (422 non-test, ~103K LOC); 16 feature modules |
| Ontology | Whelk-rs OWL 2 EL + SHACL-lite + JSON-LD validation + PROV-O provenance (PRD-022); 7 MCP ontology tools |
| Wire protocol | V4 delta is the current default (V2 = 36 B/node, V3 = 52 B/node) |
| Decision record | ~98 ADRs (ADR-011..127), plus PRDs and DDD context maps |
Documentation follows the Diátaxis framework — each quadrant serves a distinct need.
| Category | Purpose | Start here |
|---|---|---|
| Tutorials | Learning-oriented lessons that teach VisionClaw by doing | tutorials/ |
| How-to | Task recipes for deployment, development, operations, and features | how-to/ |
| Explanation | Concepts and rationale — architecture, physics, ontology, security | explanation/ |
| Reference | Exhaustive specifications — REST, WebSocket, binary protocol, schema, config | reference/ |
| Decisions | Architecture Decision Records governing every major design choice | adr/ |
| Formal record | Product Requirements, Domain-Driven Design context maps, and ADRs | prd/ · ddd/ · adr/ |
- New here? Start with What is VisionClaw?, then Installation and Your first graph.
- Building against the API? See REST API, WebSocket protocol, and Binary protocol.
- Understanding the system? Read System overview, Backend architecture, and Physics & GPU engine.
- Operating it in production? See Deployment and the operations runbooks.
Before debugging unexpected behaviour, check KNOWN_ISSUES.md — it tracks active P1/P2 bugs and their workarounds.
VisionClaw runs on top of agentbox, the sovereign agent-runtime subsystem (skills, identity mesh, pod adapters, ACSP control surfaces). agentbox is a subsystem with its own documentation set — VisionClaw links into it rather than duplicating it.
- Contributing — workflow, branching conventions, code standards
- Changelog — version history and release notes
Maintained by DreamLab AI — Issues · Discussions