Configuration Reference¶

ragwise uses Pydantic models for all configuration. Typos raise ValueError at construction time — not at first query.

RAGConfig¶

from ragwise import RAGConfig

# From environment variables
config = RAGConfig.from_env()

# From YAML file
config = RAGConfig.from_yaml("./ragwise_config.yaml")

# Inline
config = RAGConfig(
    llm=LLMConfig(provider="openai", model="gpt-4o-mini"),
    store=StoreConfig(backend="lance", path="./ragwise-index"),
)

Environment Variables¶

Env Var	RAGConfig field	Example
`RAGWISE_LLM_MODEL`	`llm.model`	`openai/gpt-4o-mini`
`RAGWISE_LLM_API_KEY`	`llm.api_key`	`sk-...`
`RAGWISE_STORE_BACKEND`	`store.backend`	`lance`, `postgresql`, `memory`
`RAGWISE_STORE_PATH`	`store.path`	`./ragwise-index`
`RAGWISE_EMBEDDER`	`embedder.model`	`openai/text-embedding-3-small`
`RAGWISE_CACHE_REDIS_URL`	`cache.redis_url`	`redis://localhost:6379`

RAG Constructor¶

from ragwise import RAG

async with RAG(
    llm="openai/gpt-4o-mini",         # or RAGConfig object
    store="memory",                    # memory | lance://path | postgresql://...
    embedder="openai/text-embedding-3-small",
    reranker="flashrank",              # flashrank | cohere/rerank-4 | None
    chunk_size=512,
    chunk_overlap=64,
    cache=True,
    cache_threshold=0.92,
    confidence_threshold=0.7,
) as rag:
    ...

Parameter	Type	Default	Description
`llm`	`str \\| LLMConfig`	required	LLM provider/model string or config
`store`	`str \\| StoreConfig`	`"memory"`	Vector store backend
`embedder`	`str \\| EmbedderConfig`	`"openai/text-embedding-3-small"`	Embedding model
`reranker`	`str \\| None`	`None`	Reranker: `"flashrank"` or `"cohere/rerank-4"`
`chunk_size`	`int`	`512`	Target chunk size in tokens
`chunk_overlap`	`int`	`64`	Overlap between consecutive chunks
`cache`	`bool`	`False`	Enable semantic query cache
`cache_threshold`	`float`	`0.92`	Cosine similarity threshold for cache hits
`confidence_threshold`	`float \\| None`	`None`	Min centroid-distance score; skips LLM if below

QueryConfig¶

from ragwise import QueryConfig

answer = await rag.query(
    "What is the refund policy?",
    config=QueryConfig(
        top_k=5,
        tenant_id="org_a",
        allowed_sources=["docs/policies/*.md"],
        as_of="2024-06-15",
        n_queries=3,
        include_citations=True,
    ),
)

Parameter	Type	Default	Description
`top_k`	`int`	`4`	Number of chunks to retrieve
`tenant_id`	`str \\| None`	`None`	Scopes retrieval to this tenant
`allowed_sources`	`list[str] \\| None`	`None`	Source glob patterns to restrict retrieval
`as_of`	`str \\| None`	`None`	ISO date; filters to chunks valid on this date
`n_queries`	`int`	`1`	Number of query variants (RAG-Fusion)
`include_citations`	`bool`	`True`	Whether to populate `answer.citations`
`cache_threshold`	`float \\| None`	`None`	Override instance-level cache threshold

Ingest Options¶

result = await rag.ingest(
    "./docs/",
    glob="**/*.md",
    tenant_id="org_a",
    metadata={"valid_from": "2024-01-01", "valid_until": "2024-12-31"},
    force=False,          # True to bypass incremental hash check
    on_progress=callback, # called per file: fn(file_path, status)
)
# result: IngestResult(chunks_created=42, skipped=3, failed_files=[])

LLM Provider Strings¶

String	Provider	Notes
`"openai/gpt-4o-mini"`	OpenAI	Requires `OPENAI_API_KEY`
`"openai/gpt-4o"`	OpenAI	—
`"anthropic/claude-haiku-4-5"`	Anthropic	Requires `ANTHROPIC_API_KEY`
`"anthropic/claude-opus-4-6"`	Anthropic	—
`"ollama/llama3"`	Ollama	Requires local Ollama server

Embedder Provider Strings¶

String	Provider	Notes
`"openai/text-embedding-3-small"`	OpenAI	Default
`"openai/text-embedding-3-large"`	OpenAI	Higher accuracy
`"local/all-MiniLM-L6-v2"`	sentence-transformers	`pip install ragwise[local-emb]`
`"local/bge-large-en-v1.5"`	sentence-transformers	Higher accuracy, larger