ckg-nvidia-nemoclaw v0.19.0

Auditable knowledge graph · NVIDIA NemoClaw · built by Graphify.md

Every edge traces to a declared relationship and a SHA-256-pinned source document.
Deterministic agent answers — no inference at query time.

55nodes

74edges

269avg tokens

0.471F1 · 4× RAG

11×fewer tokens than RAG

Connect

claude.ai connector URL:

https://ckg-nvidia-nemoclaw.onrender.com/mcp

Settings → Connectors → Add connector → paste URL.

Claude Desktop / Claude Code:

{
  "mcpServers": {
    "nemoclaw": {
      "command": "uvx",
      "args": ["ckg-nvidia-nemoclaw"]
    }
  }
}

Tools

ask_nemoclaw(question) — natural language query over the graph
query_ckg(concept, depth) — typed subgraph traversal (1–5 hops)
get_prerequisites(concept) — full upstream dependency chain
search_concepts(query) — fuzzy search across all 55 concepts
list_domains() — available domains and node/edge counts
verify_source(concept) — source URL + SHA-256 hash · full audit chain

Install

pip install ckg-nvidia-nemoclaw
# or: uvx ckg-nvidia-nemoclaw

GitHub · PyPI · Benchmark paper · Interactive graph

Code: MIT · Data: Elastic License 2.0 · Pipeline: Proprietary — Graphify.md
Community-built. Not affiliated with NVIDIA Corporation. NemoClaw is a trademark of NVIDIA Corporation.