How monal95/llm-based-security-copilot works
SecureRAG is a local RAG copilot for SOC analysts: public threat-intelligence feeds become searchable evidence, analyst queries produce grounded and claim-verified answers, and the API also exposes vulnerability triage, evidence search, MITRE details, runbooks, and evaluation results to the web UI. The core query workflow and several API calls are visible in excerpts; other product capabilities and ingestion/indexing links are documented in the README. Some UI-to-API and evaluation wiring is not shown in sampled source, so those relationships are marked unverified.
Main entry points: pipeline.py RAG orchestrator · main.py REST API · build_knowledge_base.py Chunk and embed · retriever.py Evidence search service
Components 25
Threat intelligence
- Feed ingestion
modules/ingestion
Knowledge base
- Chunk and embed
modules/Chunking/build_knowledge_base.py - Document chunking
modules/Chunking/chunker.py - Text embeddings
modules/Chunking/embedder.py - Vector index
modules/Chunking/vector_store.py - BM25 index
modules/Chunking/bm25_index.py
Grounded answers
- RAG orchestrator
modules/pipeline.py - Query expansion
modules/Retrieval/query_expander.py - Dense retrieval
modules/Retrieval/dense_retriever.py - Sparse retrieval
modules/Retrieval/sparse_retriever.py - Hybrid fusion
modules/Retrieval/hybrid_fusion.py - Evidence reranking
modules/Retrieval/reranker.py - Grounded prompting
modules/Generation/prompt_template.py - Local LLM generation
modules/Generation/llm_chain.py - Claim verification
modules/Verification/hallucination_guard.py
Analyst services
- Evidence search service
modules/retriever.py - Patch prioritization
modules/priority_scorer.py - Priority explanations
modules/patch_explainer.py - Incident runbooks
modules/runbook_generator.py - Retrieval evaluation
evaluation/evaluation_engine.py
API and analyst UI
- Analyst web views
frontend/src/components - Frontend API client
frontend/src/services/api.ts - REST API
backend/main.py
Other components
- CTI feeds
- SOC analyst
Connections 27
- CTI feeds → Feed ingestion: provides feeds
- Feed ingestion → Chunk and embed: supplies records
- Chunk and embed → Document chunking: chunks records
- Chunk and embed → Text embeddings: embeds chunks
- Chunk and embed → Vector index: builds vector index
- Chunk and embed → BM25 index: builds sparse index
- SOC analyst → Analyst web views: uses
- Analyst web views → Frontend API client: requests data
- Frontend API client → REST API: calls REST
- REST API → RAG orchestrator: runs chat
backend/main.py - REST API → Evidence search service: requests evidence
backend/main.py - REST API → Patch prioritization: scores CVEs
backend/main.py - REST API → Priority explanations: requests rationale
backend/main.py - REST API → Incident runbooks: generates runbook
backend/main.py - REST API → Retrieval evaluation: serves evaluation results
- RAG orchestrator → Query expansion: expands query
modules/pipeline.py - RAG orchestrator → Dense retrieval: retrieves dense
modules/pipeline.py - RAG orchestrator → Sparse retrieval: retrieves sparse
modules/pipeline.py - RAG orchestrator → Hybrid fusion: fuses rankings
modules/pipeline.py - RAG orchestrator → Evidence reranking: reranks evidence
modules/pipeline.py - RAG orchestrator → Grounded prompting: constructs prompt
modules/pipeline.py - RAG orchestrator → Local LLM generation: generates answer
modules/pipeline.py - RAG orchestrator → Claim verification: verifies claims
modules/pipeline.py - Evidence search service → Dense retrieval: retrieves dense
modules/retriever.py - Evidence search service → Sparse retrieval: retrieves sparse
modules/retriever.py - Evidence search service → Hybrid fusion: fuses rankings
modules/retriever.py - Evidence search service → Evidence reranking: reranks evidence
modules/retriever.py