How venkatkollu/localknowledge works
Local KnowledgeOS is a privacy-preserving, local-first document RAG system. Its main workflow is authenticated document ingestion and indexing, followed by access-controlled hybrid retrieval, evidence selection, answer generation, and citation checking. The graph also preserves index lifecycle, Git ingestion, security, observability, and evaluation capabilities described in the README. Several connections are documented rather than visible in sampled implementation excerpts; those use the README as evidence.
Main entry points: api.py FastAPI · retrieval.py Query Pipeline · ingestion.py Ingestion Pipeline · generation.py Answer Generation
Components 21
API and Jobs
- FastAPI
app/api.py - Ingestion Jobs
app/jobs.py
Document Ingestion
- Ingestion Pipeline
app/ingestion.py - Git Ingestion
app/repositories.py - Document Parsing
app/parsing.py - Structure and Chunks
app/chunking.py - Embedding Models
app/embeddings.py
Retrieval and Answers
- Query Pipeline
app/retrieval.py - Reranking
app/reranking.py - Evidence Decisions
app/evidence.py - Answer Generation
app/generation.py - Citation Validation
app/citations.py
State and Security
- Configuration
app/config.py - SQLite and FTS5
app/storage.py - Access and Screening
app/security.py - Authentication
app/auth.py
Evaluation and Tracing
- RAG Evaluation
app/evaluation.py - Retrieval Metrics
eval/metrics.py - Traces and Metrics
app/observability.py
Other components
- User
- Ollama Models
Connections 29
- User → FastAPI: submits requests
README.md - FastAPI → Authentication: authenticates
app/api.py - FastAPI → Ingestion Jobs: submits ingestion
app/api.py - Ingestion Jobs → Ingestion Pipeline: runs jobs
- FastAPI → Query Pipeline: routes queries
README.md - FastAPI → SQLite and FTS5: manages indexes
app/api.py - FastAPI → Git Ingestion: starts Git ingestion
README.md - Ingestion Pipeline → Document Parsing: parses documents
app/ingestion.py - Ingestion Pipeline → Structure and Chunks: structures chunks
app/ingestion.py - Ingestion Pipeline → Embedding Models: creates vectors
README.md - Ingestion Pipeline → Answer Generation: contextualizes chunks
app/ingestion.py - Ingestion Pipeline → SQLite and FTS5: writes index
app/ingestion.py - Git Ingestion → Ingestion Pipeline: supplies repository files
- Query Pipeline → Embedding Models: embeds queries
app/retrieval.py - Query Pipeline → SQLite and FTS5: searches index
README.md - Query Pipeline → Access and Screening: filters access
README.md - Query Pipeline → Reranking: reranks candidates
app/retrieval.py - Query Pipeline → Evidence Decisions: selects evidence
README.md - Query Pipeline → Answer Generation: requests answer
README.md - Answer Generation → Ollama Models: calls local model
README.md - Answer Generation → Citation Validation: validates answer
README.md - Evidence Decisions → Citation Validation: supports checking
app/evidence.py - FastAPI → Configuration: loads settings
app/api.py - Authentication → Access and Screening: resolves principal
app/auth.py - FastAPI → RAG Evaluation: runs evaluation
README.md - RAG Evaluation → Citation Validation: checks claims
app/evaluation.py - RAG Evaluation → Retrieval Metrics: computes measures
app/evaluation.py - FastAPI → Traces and Metrics: records traces
app/api.py - FastAPI → SQLite and FTS5: reads trace data
app/api.py