ckg-nvidia-nemoclaw 0.5.1

Auditable knowledge graph for NVIDIA NemoClaw with cryptographic source traceability — deterministic agent answers, MCP-native, 11× fewer tokens than RAG

ckg-nvidia-nemoclaw

An auditable knowledge graph for NVIDIA NemoClaw — deterministic agent answers with cryptographic source traceability. Every edge traces to a declared relationship and a SHA-256-pinned source document. Built for platform engineers, agent developers, and docs teams who need verifiable answers about dependencies, runtimes, policy, and deployment paths — not model inference.

Not a general-purpose semantic search layer — if it's not a declared edge, the graph doesn't return it.

pip install ckg-nvidia-nemoclaw

"The graph doesn't guess — it traverses. Every answer traces to a declared edge."

→ Interactive README with live graph

PyPI · GitHub · Benchmark paper · HuggingFace dataset · MCP connector URL · HN 47427027 · HN 47435066 · graphifymd.com · Pro · 97 domains · Benchmark

---

The graph

55 nodes · 74 edges · colored by layer · edge types: REQUIRES · ENABLES · IMPLEMENTS · RELATES_TO

graph TD
    NC[NemoClaw] --> OS[OpenShell]
    NC --> OC[OpenClaw]
    NC --> HM[Hermes]
    NC --> LC[LangChain Deep Agents]
    NC --> MCP[ManagedMCPServer]
    NC --> AH[AgentHeartbeat]

    OC --> PTD[ProgressiveToolDisclosure]
    HM --> PTD
    LC --> PTD

    OS --> L7[L7Proxy]
    OS --> LL[LandlockLSM]
    OS --> CP[CONNECT_Proxy]

    L7 --> SG[SharedGateway]
    SG --> IP[InferenceProvider]
    IP --> vLLM[vLLM]
    IP --> OLL[Ollama]
    IP --> NIM[NIM_Local]
    IP --> MR[ModelRouter]

    MCP --> NP[NetworkPolicy]
    MCP --> L7

    NP --> PT[PolicyTier]
    NP --> PP[PolicyPreset]
    PP --> TG[Telegram]
    PP --> DC[Discord]
    PP --> SL[Slack]

    SH[SecurityHardening] --> SB[Sandbox]
    SH --> LL
    SH --> BP[NemoClaw_Blueprint]

    NIM --> DGX[DGX_Spark]

    style NC fill:#0f6e56,color:#fff
    style PTD fill:#1a5c47,color:#fff
    style SH fill:#1a5c47,color:#fff
    style NP fill:#1a5c47,color:#fff
    style IP fill:#2d7a5e,color:#fff

---

What developers are actually hitting

Signal from 123 GitHub issues, HN 47427027, HN 47435066, and hands-on walkthroughs.

01 — Context bloat in tool loops. Agents forget tool schemas after loop iterations. The model re-infers NemoClaw's architecture on every query instead of reading declared structure.

02 — "Which agent is burning my budget?" OpenShell makes token spend visible per agent for the first time. The next question is how to reduce it. CKG is that answer.

03 — The Policy Source Gap. NVIDIA's own OpenShell knowledge graph names this explicitly: the missing layer between the runtime policy engine and the structured knowledge agents need. We filled it.

---

What it is

55 nodes · 74 edges · the full NemoClaw stack as a typed dependency graph. Pre-structured, traversable, deterministic. Served over MCP. No inference at query time.

get_prerequisites("ManagedMCPServer")

→ ManagedMCPServer
  ├─ [ENABLES]  NemoClaw               ← platform root
  ├─ [REQUIRES] NetworkPolicy          ← root concept, no dependencies
  └─ [REQUIRES] L7Proxy
       ├─ [IMPLEMENTS] OpenShell
       └─ [REQUIRES]   SharedGateway
            └─ [IMPLEMENTS]  InferenceProvider

  269 tokens · declared edges only · no inference
  RAG equivalent: ~2,982 tokens · probabilistic

query_ckg("ProgressiveToolDisclosure")

← [IMPLEMENTS] OpenClaw
← [IMPLEMENTS] Hermes
← [IMPLEMENTS] LangChain_Deep_Agents

All three runtimes share this mechanism.
RAG returns three separate docs. The graph knows — it's a declared edge.

---

Source provenance — verifiable to the byte

Every node traces to a declared source URL and a SHA-256 hash of that document's bytes at extraction time. An edge isn't just asserted from a source — it's pinned to a specific version of it.

# Verify any node's source hasn't changed since extraction
curl -s https://docs.nvidia.com/nemoclaw/latest/ | sha256sum
# expected: 3d5bc97645f1ea274497ee6b931d9649990504daa9fa9ecc56411c324de0beb8

The full audit chain:

edge answer
  → graph commit hash       (git log -- nemoclaw.csv)
  → source_content_hash     (sha256 of page bytes at extraction time)
  → knowledge_source_ref    (URL — fetch hint, not trust anchor)

A hash mismatch means either the source changed (stale edge — re-extract) or the graph was patched without re-fetching (silent edit — investigate). No judgment required. Run scripts/refresh_hashes.py to recompute.

Via MCP:

verify_source("CorporateCA")

→ source_url:  https://docs.nvidia.com/nemoclaw/latest/
  source_hash: sha256:3d5bc97645f1...
  verify:      curl -s '<url>' | sha256sum

This implements the knowledge_source_ref + source_content_hash fields from GuardrailDecisionV1 — applied here as a reference implementation for a domain CKG.

---

Declared relationships, not confidence scores

Every edge was extracted from a source document and given a type. There are no probabilistic weights, no cosine similarity scores, no confidence intervals. An edge either exists — declared, typed, sourced — or it doesn't. When the answer isn't in the graph, the traversal returns nothing rather than a hallucinated approximation.

Edge types:

Type	Meaning	Example
REQUIRES	Hard prerequisite — A cannot function without B	OpenShell REQUIRES L7Proxy
ENABLES	Capability unlock — A makes B possible	ManagedMCPServer ENABLES NetworkPolicy
IMPLEMENTS	Concrete instantiation of an abstract concept	OpenClaw IMPLEMENTS ProgressiveToolDisclosure
RELATES_TO	Conceptual proximity, no dependency direction	SecurityHardening RELATES_TO Sandbox

Why no confidence levels? Confidence scores exist to compensate for retrieval uncertainty. RAG returns chunks that might be relevant, scored by vector distance. A 0.87 cosine score doesn't tell you if the relationship is real — it tells you the embedding is similar. CKG has no retrieval step, so there's nothing to score. The edge type is the confidence signal: REQUIRES means load-bearing and sourced; RELATES_TO means real but weaker. A missing edge is not a soft no — it's silence from a source-grounded system.

✗ RAG:  "CorporateCA is probably used for identity management... (similarity: 0.81)"
        Score is on the chunk, not the claim. The claim itself is unverified.

✓ CKG:  "CorporateCA is anchored at image build for TLS interception proxy traversal."
        No score. Declared edge. Traces to security hardening source doc.

---

A/B test — NemoClaw domain, local models, no GPU

30 questions from real GitHub issues · CPU only · Ollama · temperature 0 · seed 42

Category	Bare model	+ CKG	Lift
Lookup F1	0.100	0.171	+71%
Multi-hop F1	0.058	0.100	+73%
Prereq-chain F1	0.077	0.156	+103%
Key-fact accuracy	9.3%	22.3%	+13pp

phi4-mini and nemotron-mini truncate at ~2,050 tokens. The CKG is 6,837 tokens — only 30% is loading. Prereq-chain F1 still doubles on that fraction. Full-context models widen the gap further.

L01 — lookup:

Q: What are the three agent runtimes in NemoClaw?
✗ Bare: "NemoClaw supports TensorFlow, PyTorch, and ONNX Runtime..." [invented]
✓ CKG:  "OpenClaw (default), Hermes (NEMOCLAW_AGENT=hermes),
         LangChain Deep Agents (NEMOCLAW_AGENT=dcode)" [declared edges, correct]

P08 — prereq-chain (best Δ +0.261):

Q: How does CorporateCA integrate into NemoClaw's security chain?
✗ Bare: "CorporateCA, a cloud-native IAM solution from NVIDIA..." [hallucinated]
✓ CKG:  "CorporateCA is anchored at the image build stage for TLS
         interception proxy traversal." [exact mechanism, correct]

L08 — lookup:

Q: What enterprise manufacturing deployment uses NemoClaw via the FOX Blueprint?
✗ Bare: "FOX (Flexible Open-Source Object Tracking)..." [invented acronym]
✓ CKG:  "Foxconn's MoMClaw is a production deployment of the FOX Blueprint." [correct]

Full report: ckg-ab-test/results/REPORT_nemoclaw.md

---

Install

Add to claude.ai (no install required):

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

Settings → Connectors → Add connector → paste URL.

Local — Claude Desktop / Claude Code:

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

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

---

Tools

Tool	Description
ask_nemoclaw(question)	Natural language query — auto-detects concept, traverses the relevant subgraph
query_ckg(concept, depth)	Typed subgraph around a specific concept (1–5 hops)
get_prerequisites(concept)	Full upstream prerequisite chain — every dependency in order
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 for a concept node — full audit chain back to the source bytes

---

What's in the graph

55 nodes · 74 edges · 4 edge types: REQUIRES · ENABLES · IMPLEMENTS · RELATES_TO

Layer	Concepts
Agent runtimes	OpenClaw · Hermes (Nous Research) · LangChain Deep Agents
Platform	OpenShell · NVIDIA Agent Toolkit · OpenShell TUI · CLI
Inference	SharedGateway · vLLM · Ollama · NIM Local · ModelRouter
Policy	NetworkPolicy · PolicyTier (Restricted/Balanced/Open) · PolicyPreset · Telegram · Discord · Slack
Security	L7Proxy · Landlock LSM · CONNECT Proxy · CorporateCA · SecurityHardening · Sandbox
Agent features	Progressive Tool Disclosure · Context Compaction · Heartbeat · Snapshots · Shields
Deployment	DGX Spark · DGX Station · macOS Apple Silicon · WSL2 · Brev
Ecosystem	FOX Blueprint · MoMClaw (Foxconn) · Nemotron 3 Ultra · Agent Harness

Every node traces to a source at docs.nvidia.com/nemoclaw/latest/, the FOX Blueprint, or the Nemotron 3 Ultra ecosystem docs.

---

Community pulse

123 open GitHub issues · HN 47427027 · HN 47435066

• #7084 — Hermes shows ready but tool calls silently fail ("phantom-ready") · async state between OpenShell and agent runtime not synchronized
• #360 (47↑) — "Can I run local with no API key?" · inference.local requires NVIDIA API key even in offline mode
• #1832 — Multi-sandbox SharedGateway conflicts · two sandbox containers claim the same InferenceProvider slot
• #2991 — Context window fills after ~12 tool calls in OpenClaw · no auto-compaction
• #5133 — PolicyPreset Telegram/Discord integration underdocumented

---

Sources

Every node and edge traces to one of these. No probabilistic inference — declared relationships only.

Type	Source	Coverage
Official	docs.nvidia.com/nemoclaw/latest/	Core platform — agent runtimes, OpenShell, security, inference routing, policy
Official	FOX Blueprint docs	MoMClaw (Foxconn) manufacturing deployment, enterprise patterns
Official	Nemotron 3 Ultra ecosystem	DGX Spark/Station, ModelRouter, NIM Local integration
Official	Security hardening guide	LandlockLSM, L7Proxy, CorporateCA, CONNECT Proxy, Sandbox chain
Official	Managed MCP Server docs	ManagedMCPServer, NetworkPolicy, PolicyTier, PolicyPreset, messaging bindings
Official	Agent features reference	ProgressiveToolDisclosure, Context Compaction, Heartbeat, Snapshots, Shields
Community	github.com/Yarmoluk/ckg-nvidia-nemoclaw/issues	123 issues — phantom-ready, local API key, SharedGateway conflicts
Community	HN 47427027 · HN 47435066	Local inference gap (83↑) · token burn visibility
Dataset	huggingface.co/datasets/danyarm/ckg-benchmark	KRB v0.6.2 — 7,928 queries, 30 NemoClaw-domain questions
Benchmark	github.com/Yarmoluk/ckg-benchmark/paper/main.pdf	Full methodology, F1 0.471, RAG/GraphRAG baselines
A/B report	ckg-ab-test/results/REPORT_nemoclaw.md	30-question NemoClaw run · phi4-mini + nemotron-mini · CPU

---

Benchmark (v0.6.2 locked)

System	Macro F1	Mean tokens	Cost / 1k queries
CKG	0.471	269	$7.81
RAG	0.123	2,982	$76.23
GraphRAG	0.120	~3,000	~$76

7,928 queries · 5-hop F1: 0.772 (CKG) vs 0.170 (RAG) · dataset: huggingface.co/datasets/danyarm/ckg-benchmark · full paper

---

EVAL

benchmark: ckg-benchmark v0.6.2
dataset: huggingface.co/datasets/danyarm/ckg-benchmark
benchmarked: false
rag_baseline_f1: 0.123
graphrag_baseline_f1: 0.120
mean_tokens: 269
paper: github.com/Yarmoluk/ckg-benchmark/blob/main/paper/main.pdf

---

Licensing

Layer	License	What it means
Server code — server.py, graph.py, serve.py, scripts/	MIT (LICENSE-CODE)	Use, fork, embed, modify freely — no restrictions
Graph data — domains/nemoclaw.csv	Elastic License 2.0 (LICENSE)	Free for any internal or commercial use; you may not resell this graph as a hosted/managed service
Extraction pipeline + benchmark	Proprietary — Graphify.md	Not in this repo

Can I use this in my product or agent? Yes — no restrictions. Can I build a competing "NemoClaw CKG as a Service"? No — that's what ELv2 blocks. Can I fork the server code and build my own CKG? Yes — the code is MIT.

---

Built by Graphify.md · PyPI · patent pending

Community-built. Not affiliated with, endorsed by, or sponsored by NVIDIA Corporation. NemoClaw is a trademark of NVIDIA Corporation. All referenced trademarks belong to their respective owners.