Configuration

Configuration section: ai.rag (in application.yaml) or env prefix LEX_AI_RAG__.

The provider’s config_key = "ai.rag" — a nested dot-separated key. The orchestrator reads ai.rag: from YAML and coerces it into RAGConfig.

RAGConfig

Top-level configuration. Full example:

ai.rag:
  enabled: true
  vector_store_type: pgvector
  vector_dimension: 1536
  collection_name: docs
  top_k: 5
  similarity_threshold: 0.7
  use_hybrid_search: true
  embedding_provider: openai
  embedding_model: text-embedding-3-small
  enable_citations: true
  citation_style: numbered
  chunk_size: 512
  chunk_overlap: 50
  chunking_strategy: recursive
  enable_query_expansion: true
  enable_hyde: false
  synthesis_strategy: hybrid
  enable_hallucination_detection: true
  enable_caching: true
  cache_ttl: 3600

Env-var override form (nested keys use __):

export LEX_AI_RAG__VECTOR_STORE_TYPE=qdrant
export LEX_AI_RAG__TOP_K=10
export LEX_AI_RAG__EMBEDDING_MODEL=text-embedding-3-large

Field Reference

Key	Type	Default	Env Var	Description
`enabled`	`bool`	`True`	`LEX_AI_RAG__ENABLED`	Enable the RAG pipeline
`persist_directory`	`str \| None`	`None`	`LEX_AI_RAG__PERSIST_DIRECTORY`	Local directory for file-based stores (Chroma)
`vector_store_type`	`str`	`"pgvector"`	`LEX_AI_RAG__VECTOR_STORE_TYPE`	Backend: `pgvector`, `chroma`, `qdrant`, `mock`
`vector_dimension`	`int`	`1536`	`LEX_AI_RAG__VECTOR_DIMENSION`	Embedding dimension (1536 for ada-002)
`collection_name`	`str`	`"default"`	`LEX_AI_RAG__COLLECTION_NAME`	Collection/index name
`top_k`	`int`	`5`	`LEX_AI_RAG__TOP_K`	Number of documents to retrieve
`similarity_threshold`	`float`	`0.7`	`LEX_AI_RAG__SIMILARITY_THRESHOLD`	Minimum similarity score (0–1)
`use_hybrid_search`	`bool`	`True`	`LEX_AI_RAG__USE_HYBRID_SEARCH`	Enable hybrid (semantic + keyword)
`embedding_provider`	`str`	`"openai"`	`LEX_AI_RAG__EMBEDDING_PROVIDER`	Provider: `openai`, `cohere`, etc.
`embedding_model`	`str \| None`	`None`	`LEX_AI_RAG__EMBEDDING_MODEL`	Must be set explicitly
`enable_citations`	`bool`	`True`	`LEX_AI_RAG__ENABLE_CITATIONS`	Include source citations in responses
`citation_style`	`str`	`"inline"`	`LEX_AI_RAG__CITATION_STYLE`	`inline`, `footnote`, `numbered`
`min_citation_confidence`	`float`	`0.6`	`LEX_AI_RAG__MIN_CITATION_CONFIDENCE`	Min confidence for citations (0–1)
`chunk_size`	`int`	`512`	`LEX_AI_RAG__CHUNK_SIZE`	Text chunk size in tokens
`chunk_overlap`	`int`	`50`	`LEX_AI_RAG__CHUNK_OVERLAP`	Overlap between consecutive chunks
`chunking_strategy`	`str`	`"recursive"`	`LEX_AI_RAG__CHUNKING_STRATEGY`	`recursive`, `semantic`, `token`
`enable_query_expansion`	`bool`	`True`	`LEX_AI_RAG__ENABLE_QUERY_EXPANSION`	Enable query expansion
`enable_hyde`	`bool`	`False`	`LEX_AI_RAG__ENABLE_HYDE`	Enable HyDE
`synthesis_strategy`	`str`	`"hybrid"`	`LEX_AI_RAG__SYNTHESIS_STRATEGY`	`direct`, `extractive`, `abstractive`, `hybrid`
`enable_hallucination_detection`	`bool`	`True`	`LEX_AI_RAG__ENABLE_HALLUCINATION_DETECTION`	Check for hallucinations
`enable_caching`	`bool`	`True`	`LEX_AI_RAG__ENABLE_CACHING`	Cache RAG query results
`cache_ttl`	`int`	`3600`	`LEX_AI_RAG__CACHE_TTL`	Cache TTL in seconds

PipelineConfig

from lexigram.ai.rag.config import PipelineConfig

config = PipelineConfig(
    name="my-pipeline",
    stages=["RETRIEVAL", "SYNTHESIS", "QUALITY_ASSURANCE"],
    max_retries=3,
    require_citations=True,
    auto_evaluate_every_n=100,
)

Key	Type	Default	Description
`name`	`str`	`"default-rag-pipeline"`	Pipeline identifier
`stages`	`list[PipelineStageType]`	`[RETRIEVAL, SYNTHESIS, QUALITY_ASSURANCE]`	Ordered pipeline stages
`max_retries`	`int`	`3`	Retries per stage on transient failure
`retry_delay`	`float`	`1.0`	Delay between retries (seconds)
`default_error_strategy`	`str`	`"graceful"`	`graceful`, `fail_fast`, `skip`
`require_citations`	`bool`	`False`	Raise `MissingCitationsError` if no citations
`auto_evaluate_every_n`	`int \| None`	`None`	Auto-evaluation frequency

Sub-configs (ingestion, query_processing, retrieval, context_optimization, synthesis, quality_assurance, post_processing) each expose their own fields. See lexigram.ai.rag.config for the full schema.

Environment Variables

All keys map from env vars using LEX_AI_RAG__ prefix with __ as nested delimiter:

LEX_AI_RAG__ENABLED=true
LEX_AI_RAG__VECTOR_STORE_TYPE=qdrant
LEX_AI_RAG__EMBEDDING_MODEL=text-embedding-3-small
LEX_AI_RAG__CHUNK_SIZE=1024
LEX_AI_RAG__USE_HYBRID_SEARCH=false