Memory Modules

This page documents the stable public memory facade.

Stable public memory facade exports.

class design_research_agents.memory.EmbeddingProvider(*args, **kwargs)

Protocol for converting text payloads into dense vectors.

embed(texts)

Embed one or more texts.

Parameters:

texts – Input texts to embed.

Returns:

Embedding vectors in input order, or None when embeddings are unavailable and lexical fallback should be used.

property model_name

Return stable embedding model identifier.

Returns:

Stable embedding model identifier.

class design_research_agents.memory.LLMEmbeddingProvider(*, llm_client, model=None, enable_lexical_fallback=True)

Embedding adapter that delegates to an LLM client/backend when available.

Initialize one embedding adapter from an existing LLM client.

Parameters:

• llm_client – LLM client instance that may expose embeddings.

• model – Optional explicit embedding model identifier.

• enable_lexical_fallback – Return None instead of raising when embeddings are unsupported.

embed(texts)

Embed texts through available client/backend embedding APIs.

Parameters:

texts – Input texts to embed.

Returns:

List of vectors, or None when lexical fallback is enabled and embeddings are unavailable.

Raises:

Exception – Raised when embeddings fail and lexical fallback is disabled.

property model_name

Return configured embedding model identifier.

Returns:

Resolved embedding model identifier.

class design_research_agents.memory.SQLiteMemoryStore(*, db_path=None, embedding_provider=None, embedding_model=None)

Persistent local memory store with optional embedding-based retrieval.

Initialize a local sqlite memory store.

Parameters:

• db_path – Optional sqlite database file path.

• embedding_provider – Optional embedding provider for vector retrieval.

• embedding_model – Optional model key used for embedding rows.

close()

Close sqlite connection.

property db_path

Return sqlite database path used by this store.

Returns:

Configured sqlite database path.

search(query)

Search records using lexical relevance and optional vector similarity.

Parameters:

query – Structured memory search query.

Returns:

Ranked memory records matching the query.

write(records, *, namespace='default')

Persist records and return normalized stored records.

Parameters:

• records – Memory record payloads to persist.

• namespace – Namespace partition for writes.

Returns:

Stored records in caller-provided order.