How aditya25-yadav/enterprise-knowledge-agent works
Enterprise Knowledge Agent ingests content from six enterprise systems, normalizes it into OKF, and builds a cross-connector catalog plus vector, keyword, and developer-graph indexes. The principal user workflow is query scope parsing, planner-driven retrieval, evidence reranking and evaluation, and citation-ready answer generation; the planner can reformulate queries when evidence is incomplete. RBAC constrains retrieval, while checkpoints preserve multi-turn state. Edges follow the excerpts, SOURCE INDEX references, or README; unsampled subsystem wiring is left unasserted.
Main entry points: langgraph_planner.py LangGraph planner · pipeline.py Ingestion pipeline · retrieval Search suite · connectors Enterprise connectors
Components 26
Ingestion and indexing
- Enterprise connectors
backend/connectors - OKF schema
backend/models/okf.py - Global catalog
backend/ingestion/catalog_aggregator.py - Structure chunker
backend/ingestion/chunker.py - Knowledge chunks
backend/ingestion/chunk.py - Local embedder
backend/ingestion/embedder.py - Ingestion pipeline
backend/ingestion/pipeline.py - Vector index
backend/storage/qdrant_client.py - BM25 index
backend/storage/bm25_index.py
Retrieval and security
- Retrieval tools
backend/agent/tools.py - Search suite
backend/retrieval - RBAC enforcement
backend/security/rbac_resolver.py - Cross-encoder reranker
backend/ranking/reranker.py - Entity graph search
backend/retrieval/entity_graph.py
Agent and answers
- Query scope
backend/agent/query_scope.py - LangGraph planner
backend/agent/langgraph_planner.py - Evidence evaluator
backend/evaluation/evaluator.py - Query reformulator
backend/agent/reformulator.py - Evidence context
backend/generation/context_builder.py - Answer generator
backend/generation/answer_generator.py
Developer graph
- GitHub graph extractor
backend/graph/github_extractor.py - Developer graph store
backend/graph/neo4j_client.py
Providers and persistence
- LLM provider factory
backend/llm/factory.py - Gemini and Ollama
backend/llm - Session checkpoints
backend/storage/checkpointers.py
Other components
- User
Connections 29
- Enterprise connectors → OKF schema: normalize into
- Enterprise connectors → Global catalog: catalog assets
- OKF schema → Ingestion pipeline: ingest concepts
backend/ingestion/pipeline.py - Ingestion pipeline → Structure chunker: chunks concepts
backend/ingestion/pipeline.py - Structure chunker → Knowledge chunks: creates chunks
backend/ingestion/chunker.py - Ingestion pipeline → Local embedder: embeds chunks
backend/ingestion/pipeline.py - Ingestion pipeline → Vector index: upserts vectors
backend/ingestion/pipeline.py - Ingestion pipeline → BM25 index: indexes chunks
backend/ingestion/pipeline.py - User → Query scope: submits query
backend/agent/query_scope.py - Query scope → LangGraph planner: routes scoped query
backend/agent/langgraph_planner.py - LangGraph planner → Retrieval tools: dispatches tools
backend/agent/langgraph_planner.py - Retrieval tools → Search suite: invokes searches
backend/agent/tools.py - Search suite → Vector index: searches vectors
- Search suite → BM25 index: searches keywords
- Search suite → RBAC enforcement: filters by permissions
- Vector index → RBAC enforcement: builds RBAC filter
backend/storage/qdrant_client.py - LangGraph planner → Cross-encoder reranker: reranks evidence
backend/agent/langgraph_planner.py - LangGraph planner → Evidence evaluator: evaluates evidence
backend/agent/langgraph_planner.py - LangGraph planner → Query reformulator: reformulates query
backend/agent/langgraph_planner.py - LangGraph planner → Evidence context: builds context
backend/agent/langgraph_planner.py - Answer generator → Evidence context: formats evidence
backend/generation/answer_generator.py - LangGraph planner → LLM provider factory: requests tool reasoning
backend/agent/langgraph_planner.py - Answer generator → LLM provider factory: requests synthesis
backend/generation/answer_generator.py - LLM provider factory → Gemini and Ollama: selects provider
backend/llm/factory.py - LangGraph planner → Session checkpoints: persists sessions
backend/agent/langgraph_planner.py - Enterprise connectors → GitHub graph extractor: supplies GitHub data
backend/graph/github_extractor.py - GitHub graph extractor → Developer graph store: writes graph bundle
backend/graph/neo4j_client.py - Retrieval tools → Entity graph search: dispatches entity search
backend/agent/tools.py - Entity graph search → Developer graph store: traverses graph
backend/retrieval/entity_graph.py