redis_openai_agents.middleware.ConversationMemoryMiddleware#

class ConversationMemoryMiddleware(history, *, session_tag=None, top_k=5, distance_threshold=None, persist_reply=True, response_text_extractor=None)[source]#

Prepend semantically relevant past messages into the request.

Parameters:

  • history (SemanticMessageHistory) – A SemanticMessageHistory instance backing the retrieval and storage. Callers are responsible for its lifecycle.

  • session_tag (str | None) – Tag passed to history queries and inserts. Allows tenant / user / conversation isolation.

  • top_k (int) – Maximum number of past messages to prepend.

  • distance_threshold (float | None) – Optional override for relevance matching.

  • persist_reply (bool) – When True (default), the assistant reply is also stored back into history so follow-up turns can retrieve it.

  • response_text_extractor (Any) – Optional callable that turns a model response into the text to store. Defaults to best-effort extraction from OpenAI Responses-shaped responses.

__init__(history, *, session_tag=None, top_k=5, distance_threshold=None, persist_reply=True, response_text_extractor=None)[source]#
Parameters:

  • history (SemanticMessageHistory)

  • session_tag (str | None)

  • top_k (int)

  • distance_threshold (float | None)

  • persist_reply (bool)

  • response_text_extractor (Any)

Return type:

None

Methods

__init__(history, *[, session_tag, top_k, ...])

awrap_model_call(request, handler)