NVIDIA NemoClaw as a traversable knowledge graph — sandboxed AI agent infrastructure, OpenShell runtime, inference routing, and network policy — MCP-native
Project description
ckg-nvidia-nemoclaw
NVIDIA NemoClaw as a traversable knowledge graph — MCP-native.
55 nodes · 74 edges covering the full NemoClaw stack: OpenShell runtime, OpenClaw, Hermes, LangChain Deep Agents, inference routing, network policy, security architecture (L7 proxy, Landlock LSM, CONNECT proxy), progressive tool disclosure, managed MCP servers, deployment paths (local, Brev), FOX Blueprint, and the Nemotron 3 Ultra agent harness ecosystem.
Instead of retrieving docs, the agent traverses declared typed relationships. Every answer traces to a declared edge — the graph doesn't guess, it traverses.
Benchmark (KRB v0.6.2 locked): CKG F1 0.471 · RAG baseline 0.123 · GraphRAG baseline 0.120 · 269 tokens/query vs 2,982 (11× reduction)
Install
pip install ckg-nvidia-nemoclaw
Run as MCP Server
uvx ckg-nvidia-nemoclaw
Claude Desktop Config
{
"mcpServers": {
"nemoclaw": {
"command": "uvx",
"args": ["ckg-nvidia-nemoclaw"]
}
}
}
Tools
| Tool | Description |
|---|---|
ask_nemoclaw(question) |
Auto-detect and traverse the relevant subgraph |
query_ckg(concept, depth) |
Typed subgraph around a specific concept |
get_prerequisites(concept) |
Full upstream prerequisite chain |
search_concepts(query) |
Fuzzy search concepts by name or keyword |
list_domains() |
List available domains and node/edge counts |
Coverage
- Agent runtimes: OpenClaw · Hermes (Nous Research) · LangChain Deep Agents Code
- Platform: OpenShell · NVIDIA Agent Toolkit · CONNECT proxy · L7 proxy · Landlock LSM
- Inference: inference.local routing · SharedGateway · vLLM · Ollama · Local NIM · ModelRouter
- Policy: NetworkPolicy · PolicyTier (Restricted/Balanced/Open) · PolicyPreset bundles
- Security: 5-layer architecture (Network/Filesystem/Process/Gateway Auth/Inference)
- Features: Progressive Tool Disclosure · Context Compaction · Managed MCP Servers · Snapshots · Shields
- Deployment: Local CLI · Brev CLI · Brev Web UI · DGX Spark · macOS Apple Silicon · WSL2
- Ecosystem: FOX Blueprint · MoMClaw (Foxconn) · Nemotron 3 Ultra · Agent Harness ecosystem
Benchmark
github.com/Yarmoluk/ckg-benchmark/blob/main/paper/main.pdf
Built by Graphify.md · patent pending
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file ckg_nvidia_nemoclaw-0.2.0.tar.gz.
File metadata
- Download URL: ckg_nvidia_nemoclaw-0.2.0.tar.gz
- Upload date:
- Size: 6.7 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
00bcce727613775d83bdd0590e2061c49746258ba1a463c6a0984ad80070d71a
|
|
| MD5 |
e71ebe07d2a27bb7894ab259eb65299b
|
|
| BLAKE2b-256 |
e40c89da0482f3a68ae7ef928f817d4a606aceb5621d6403d3b55794ea9ade3d
|
File details
Details for the file ckg_nvidia_nemoclaw-0.2.0-py3-none-any.whl.
File metadata
- Download URL: ckg_nvidia_nemoclaw-0.2.0-py3-none-any.whl
- Upload date:
- Size: 9.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.14
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
f1179921cbe1d96f451bc17874b248a57cd83cc69079c33a97385cb8b22fdb5a
|
|
| MD5 |
24be50b5588ef7c94d914f7c3be18e25
|
|
| BLAKE2b-256 |
a607f7af9be7e45f50ec11fce90f40dc98fabb0a7fc1c837a276a8dc06d9398e
|