How wandersonamello/graphrag-agente-auditoria works
This repository provides a GraphRAG system for exploring Brazilian public-audit findings: a Python agent answers questions using a knowledge graph and optional semantic search. A separate pipeline derives structural network analysis from entity JSON. The main workflow is question → agent tool selection → graph or semantic retrieval → model-generated response. The README is empty and source excerpts are partial, so the graph includes only documented capabilities and supported connections; wiring for some GUI and pipeline exports is not detailed.
Main entry points: agent_graphrag.py Agno agent · embeddings.py Embedding index · graph_index.py Knowledge graph index · graphrag_pipeline.py Network analysis pipeline
Components 17
Question answering
- Agno agent
agent_graphrag.py - CLI chat
agent_graphrag.py - AgentOS / AGUI
agent_graphrag.py - Chat model
Retrieval indexes
- Knowledge graph index
graph_index.py - Graph retrieval tools
graph_index.py - Embedding index
embeddings.py - Google embedding API
Structural analysis
- Network analysis pipeline
Grafo/graphrag_pipeline.py - Entity normalization
Grafo/graphrag_pipeline.py - Community analysis
Grafo/graphrag_pipeline.py - Network visualizations
Grafo/graphrag_pipeline.py
Data and state
- Entity records
GraphRAG/resultado_entidades_final.json - Embedding cache
GraphRAG/resultado_entidades_final_embeddings_cache.json - Chat session store
agent_graphrag.py - Structural results
Grafo/graphrag_resultado.json
Other components
- Questioner
Connections 19
- Questioner → CLI chat: enters query
- CLI chat → Agno agent: runs prompt
- Agno agent → Graph retrieval tools: invokes graph tools
- Graph retrieval tools → Knowledge graph index: searches graph
- Agno agent → Embedding index: invokes semantic search
- Agno agent → Chat model: requests response
- Agno agent → Chat session store: reads/writes history
- Embedding index → Entity records: builds records from
- Embedding index → Embedding cache: loads/saves
- Embedding index → Google embedding API: embeds text and query
- Knowledge graph index → Entity records: loads records from
- Knowledge graph index → Structural results: optionally loads metrics
- Questioner → AgentOS / AGUI: uses GUI
- AgentOS / AGUI → Agno agent: serves agent
- Entity records → Network analysis pipeline: provides findings
- Network analysis pipeline → Entity normalization: normalizes and classifies
- Entity normalization → Community analysis: builds analysis graph
- Community analysis → Network visualizations: produces network views
- Network analysis pipeline → Structural results: exports analysis data