erissatallan/OpenClaw-NemoClaw-Interactive-Graph
A living knowledge graph of the OpenClaw and NVIDIA NemoClaw open-source ecosystems, extensible to community forks.
Platform-specific configuration:
{
"mcpServers": {
"OpenClaw-NemoClaw-Interactive-Graph": {
"command": "npx",
"args": [
"-y",
"OpenClaw-NemoClaw-Interactive-Graph"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
AI-powered knowledge graph builder and RAG system for the OpenClaw and NemoClaw ecosystems
[](https://python.org) [](LICENSE)
ClawGraph automatically crawls open-source repositories, extracts entities and relationships into a knowledge graph, and serves grounded answers via RAG — all secured with a 5-layer prompt injection defense. The interface is a Telegram app bot that uses the OpenClaw skill to interact with the knowledge graph.
| Conceptual Component | Implementation | |---|---| | Orchestration Pipeline | Directed Acyclic Graph-based crawl→extract→embed→graph→curate pipeline with retry and scheduling | | Custom MCP Server | Python MCP server exposing 9 GitHub API tools via the Model Context Protocol | | AI Pipeline | Gemini Flash Lite (extraction/classification) + Gemini 2.5 Flash (reasoning/curation) | | Knowledge Graph & RAG | Neo4j / NetworkX graph + embedding search → graph-grounded answer generation with Chain of Thought reasoning | | Prompt Injection Prevention | 5-layer defense: sanitizer → classifier → canary tokens → output guardrails → audit |
GitHub Repos ──▶ [GitHub MCP Server] ──▶ [Orchestration Pipeline] ──▶ [Knowledge Graph]
(Python) crawl→extract→embed→ (Neo4j / NetworkX)
graph_update→curate │
▼
Telegram Bot ◀── [OpenClaw Skill] ◀── [RAG Engine] ◀── [Graph + Vector Retrieval]
(Gemini 2.5 Flash + CoT)
│
[5-Layer Security Defense]
L1: Input Sanitizer
Loading reviews...