rag.core.config

Configuration for the papers RAG pipeline.

Attributes

PAPERS_DIR_ENV
CHROMA_DIR_ENV
CHROMA_COLLECTION_ENV
EMBED_MODEL_ENV
OPENAI_API_KEY_ENV
OPENAI_MODEL_ENV
CHUNK_SIZE_ENV
CHUNK_OVERLAP_ENV
SIMILARITY_TOP_K_ENV
DEFAULT_COLLECTION_NAME
DEFAULT_EMBED_MODEL
DEFAULT_LLM_MODEL
DEFAULT_CHUNK_SIZE
DEFAULT_CHUNK_OVERLAP
DEFAULT_SIMILARITY_TOP_K

Classes

RagConfig

Runtime settings for ingest + query.

Functions

find_repo_root(→ pathlib.Path)	Walk parents until the monorepo root (contains libs/rag-core) is found.
load_repo_dotenv(→ None)	Load .env from cwd and the monorepo root.

Module Contents

rag.core.config.PAPERS_DIR_ENV: Final = 'PAPERS_DIR'

rag.core.config.CHROMA_DIR_ENV: Final = 'CHROMA_DIR'

rag.core.config.CHROMA_COLLECTION_ENV: Final = 'CHROMA_COLLECTION'

rag.core.config.EMBED_MODEL_ENV: Final = 'EMBED_MODEL'

rag.core.config.OPENAI_API_KEY_ENV: Final = 'OPENAI_API_KEY'

rag.core.config.OPENAI_MODEL_ENV: Final = 'OPENAI_MODEL'

rag.core.config.CHUNK_SIZE_ENV: Final = 'CHUNK_SIZE'

rag.core.config.CHUNK_OVERLAP_ENV: Final = 'CHUNK_OVERLAP'

rag.core.config.SIMILARITY_TOP_K_ENV: Final = 'SIMILARITY_TOP_K'

rag.core.config.DEFAULT_COLLECTION_NAME: Final = 'papers'

rag.core.config.DEFAULT_EMBED_MODEL: Final = 'BAAI/bge-small-en-v1.5'

rag.core.config.DEFAULT_LLM_MODEL: Final = 'gpt-4o-mini'

rag.core.config.DEFAULT_CHUNK_SIZE: Final = 1024

rag.core.config.DEFAULT_CHUNK_OVERLAP: Final = 128

rag.core.config.DEFAULT_SIMILARITY_TOP_K: Final = 5

rag.core.config.find_repo_root(*, start: pathlib.Path | None = None) → pathlib.Path

Walk parents until the monorepo root (contains libs/rag-core) is found.

Parameters:

start (Path or None) – Directory to start from. Defaults to this file’s location.

Returns:

Absolute path to the monorepo root.

Return type:

Path

Raises:

FileNotFoundError – If no ancestor contains libs/rag-core.

rag.core.config.load_repo_dotenv() → None

Load .env from cwd and the monorepo root.

Existing process environment wins (override=False). Streamlit sets cwd to the script directory, so a repo-root .env is missed by a bare load_dotenv().

class rag.core.config.RagConfig

Runtime settings for ingest + query.

Variables:

• papers_dir (Path) – Directory of PDF papers to ingest.

• chroma_dir (Path) – Persistent Chroma directory (created if missing).

• collection_name (str) – Chroma collection name.

• embed_model_name (str) – HuggingFace embedding model id (local).

• llm_model_name (str) – OpenAI chat model id.

• chunk_size (int) – SentenceSplitter chunk size.

• chunk_overlap (int) – SentenceSplitter overlap.

• similarity_top_k (int) – Retrieval top-k for the query engine.

• strategy (RagStrategy) – In-process orchestration backend (LlamaIndex or LangChain).

papers_dir: pathlib.Path

chroma_dir: pathlib.Path

collection_name: str = 'papers'

embed_model_name: str = 'BAAI/bge-small-en-v1.5'

llm_model_name: str = 'gpt-4o-mini'

chunk_size: int = 1024

chunk_overlap: int = 128

similarity_top_k: int = 5

strategy: rag.core.strategy.RagStrategy

classmethod from_env(*, repo_root: pathlib.Path | None = None, strategy: rag.core.strategy.RagStrategy | None = None) → RagConfig

Build config from environment variables with repo-relative defaults.

Parameters:

• repo_root (Path or None) – Monorepo root. Discovered automatically when omitted.

• strategy (RagStrategy or None) – Orchestration backend. Defaults to RAG_STRATEGY or LlamaIndex.

Returns:

Resolved configuration.

Return type:

RagConfig

ensure_dirs() → None

Create the Chroma persistence directory if needed.