Skip to main content

Adapters API Reference

Adapters are pluggable interfaces for integrating rag_control with external services. Import from rag_control.adapters.

LLM​

Handles text generation from prompts. Types: ChatMessage = dict[str, str], PromptInput = str | list[ChatMessage]

Methods:

  • generate(prompt: PromptInput, temperature: float | None = None, max_output_tokens: int | None = None, user_context: UserContext | None = None) -> LLMResponse
  • stream(prompt: PromptInput, temperature: float | None = None, max_output_tokens: int | None = None, user_context: UserContext | None = None) -> LLMStreamResponse

QueryEmbedding​

Converts query text to embedding vectors.

Properties:

  • embedding_model: str - Canonical embedding model identifier

Methods:

  • embed(query: str, user_context: UserContext | None = None) -> QueryEmbeddingResponse

VectorStore​

Searches for documents similar to a query embedding.

Properties:

  • embedding_model: str - Canonical embedding model identifier expected by this index

Methods:

  • search(embedding: list[float], top_k: int = 5, user_context: UserContext | None = None, filter: Filter | None = None) -> VectorStoreSearchResponse

Return Types​

LLMResponse​

  • content: str - Generated text
  • usage: LLMUsage - Token counts (prompt_tokens, completion_tokens, total_tokens)
  • metadata: LLMMetadata - Model, provider, latency_ms, request_id, timestamp, temperature, top_p, raw

LLMStreamResponse​

  • stream: Iterator[LLMStreamChunk] - Stream of chunks (delta: str)
  • usage: LLMUsage | None - Token counts
  • metadata: LLMMetadata | None - Metadata

QueryEmbeddingResponse​

  • embedding: list[float] - Embedding vector
  • metadata: QueryEmbeddingMetadata - Model, provider, latency_ms, dimensions, request_id, timestamp, raw

VectorStoreSearchResponse​

  • records: list[VectorStoreRecord] - Retrieved documents (id, content, score, metadata)
  • metadata: VectorStoreSearchMetadata - Provider, index, latency_ms, top_k, returned, request_id, timestamp, raw

Implementation Examples​

See Adapters Concept Guide for detailed implementation examples with OpenAI, Anthropic, Pinecone, and other services.

See Also​