How arihntk/raggit works
raggit is a document-indexing and retrieval-augmented question-answering system. Users submit documents through the CLI, HTTP API, watcher, or optional MCP interface; ingestion parses, chunks, cleans, protects, embeds, and stores them. Queries are sanitized and optionally rewritten, searched with hybrid retrieval, fused and reranked, then augmented into cited answers with safety checks. PostgreSQL and Qdrant hold metadata and vectors; storage adapters and LLM providers supply external services. The README documents the main pipelines; source-index references support selected internal wiring, while some API, MCP, and evaluation details are not sampled.
Main entry points: indexer.py Indexing orchestrator · engine.py Query retrieval engine · main.py CLI · server.py HTTP API
Components 23
User interfaces
- CLI
raggit/cli/main.py - HTTP API
raggit/api/server.py - MCP server
raggit/mcp_server.py - Storage watcher
raggit/core/watcher.py
Document ingestion
- Indexing orchestrator
raggit/ingestion/indexer.py - Storage adapters
raggit/storage/factory.py - Format parser
raggit/ingestion/parser.py - Chunking and cleaning
raggit/ingestion/chunker.py - PII and injection safeguards
raggit/ingestion/injection.py - Embedder
raggit/ingestion/embedder.py
Retrieval and answers
- Query retrieval engine
raggit/retrieval/engine.py - Query rewriting and sanitizing
raggit/retrieval/rewrite.py - Rank fusion and reranking
raggit/retrieval/rrf.py - Answer augmentation
raggit/llm/augmenter.py - LLM providers
raggit/llm/factory.py
State and integrations
- Shared domain models
raggit/api/models.py - Audit logging
raggit/core/audit.py - PostgreSQL
- Qdrant vector store
raggit/db/vector.py
Evaluation
- Evaluation framework
raggit/eval/pipeline.py - Evaluation models
raggit/eval/models.py
Other components
- User
- Qdrant service
Connections 35
- User → CLI: runs commands
- User → HTTP API: sends requests
- User → MCP server: connects optionally
- Storage watcher → Storage adapters: watches sources
raggit/core/watcher.py - Storage watcher → Indexing orchestrator: triggers indexing
raggit/core/watcher.py - CLI → Indexing orchestrator: starts ingestion
- HTTP API → Indexing orchestrator: triggers ingestion
raggit/api/server.py - MCP server → Indexing orchestrator: offers ingestion
raggit/mcp_server.py - Indexing orchestrator → Storage adapters: reads documents
raggit/ingestion/indexer.py - Indexing orchestrator → Format parser: parses files
raggit/ingestion/indexer.py - Indexing orchestrator → Chunking and cleaning: chunks and cleans
raggit/ingestion/indexer.py - Indexing orchestrator → PII and injection safeguards: hardens content
raggit/ingestion/indexer.py - Indexing orchestrator → Embedder: embeds chunks
raggit/ingestion/indexer.py - Indexing orchestrator → PostgreSQL: stores metadata
- Indexing orchestrator → Qdrant vector store: writes vectors
raggit/ingestion/indexer.py - Qdrant vector store → Qdrant service: persists vectors
raggit/db/vector.py - CLI → Query retrieval engine: runs queries
- HTTP API → Query retrieval engine: serves queries
raggit/api/server.py - MCP server → Query retrieval engine: offers query tool
raggit/mcp_server.py - Query retrieval engine → Query rewriting and sanitizing: sanitizes and rewrites
raggit/retrieval/engine.py - Query retrieval engine → PostgreSQL: searches BM25
- Query retrieval engine → Qdrant vector store: searches semantically
raggit/retrieval/engine.py - Query retrieval engine → Rank fusion and reranking: fuses and reranks
raggit/retrieval/engine.py - Query retrieval engine → Answer augmentation: provides retrieved context
raggit/cli/query_cmd.py - Answer augmentation → LLM providers: generates answer
raggit/llm/augmenter.py - Answer augmentation → PII and injection safeguards: hardens context
raggit/llm/augmenter.py - Answer augmentation → Query retrieval engine: checks groundedness
- Indexing orchestrator → Audit logging: audits ingestion
raggit/ingestion/indexer.py - Query retrieval engine → Audit logging: audits queries
raggit/retrieval/engine.py - Audit logging → PostgreSQL: writes audit events
raggit/core/audit.py - CLI → Evaluation framework: runs evaluations
raggit/cli/main.py - HTTP API → Evaluation framework: exposes evaluations
raggit/api/server.py - Evaluation framework → Evaluation models: scores datasets
raggit/eval/pipeline.py - Evaluation framework → Query retrieval engine: evaluates retrieval
raggit/eval/pipeline.py - Shared domain models → Query retrieval engine: defines query results
raggit/retrieval/engine.py