Summary
Add a ValkeyMemoryService implementation for the ADK community memory module, backed by Valkey using the Valkey Search module for vector similarity search (HNSW). This would use the valkey-glide client library and provide functionality analogous to VertexAiRagMemoryService for developers with Valkey infrastructure.
Motivation
Currently, the community memory module has limited backend options. Valkey (the open-source fork of Redis) with the Search module provides a high-performance, self-hosted vector database option that many teams already run in production. Adding a Valkey-backed memory service would expand the ecosystem for developers who prefer or already use Valkey infrastructure.
Proposed Implementation
ValkeyMemoryService class implementing BaseMemoryService interfaceValkeyMemoryServiceConfig (Pydantic) with configurable: similarity_top_k, vector_distance_threshold, embedding_dimensions, key_prefix, index_name, distance_metric (COSINE/L2/IP), ttl_seconds- Accepts a configurable async embedding function (users bring their own embedder — OpenAI, Gemini, sentence-transformers, etc.)
- Stores memories as Valkey Hash keys with FLOAT32 vector embeddings
- Uses
FT.CREATE with VECTOR field (HNSW algorithm) + TAG fields for app_name/user_id scoped queries
- Uses
FT.SEARCH with KNN for vector similarity retrieval, pre-filtered by TAG
- Implements both
add_session_to_memory and add_events_to_memory
- Uses
Batch pipelining for efficient ingestion (1 round trip regardless of event count)
asyncio.Lock with double-check locking for thread-safe index creation- Configurable distance threshold, distance metric, TTL
Dependencies
valkey-glide >= 2.4.0 (as optional dependency under [valkey] extra)- Valkey server with the Search module loaded (e.g.,
valkey/valkey-bundle image)
Testing Plan
- Unit tests (mocked client): cover all public methods, edge cases, error conditions
- Integration tests: run against live Valkey 9.1 (
valkey/valkey-bundle image) covering vector similarity search, user/app isolation, top-k limits, distance threshold filtering, metadata preservation, index idempotency
Summary
Add a
ValkeyMemoryServiceimplementation for the ADK community memory module, backed by Valkey using the Valkey Search module for vector similarity search (HNSW). This would use thevalkey-glideclient library and provide functionality analogous toVertexAiRagMemoryServicefor developers with Valkey infrastructure.Motivation
Currently, the community memory module has limited backend options. Valkey (the open-source fork of Redis) with the Search module provides a high-performance, self-hosted vector database option that many teams already run in production. Adding a Valkey-backed memory service would expand the ecosystem for developers who prefer or already use Valkey infrastructure.
Proposed Implementation
ValkeyMemoryServiceclass implementingBaseMemoryServiceinterfaceValkeyMemoryServiceConfig(Pydantic) with configurable:similarity_top_k,vector_distance_threshold,embedding_dimensions,key_prefix,index_name,distance_metric(COSINE/L2/IP),ttl_secondsFT.CREATEwith VECTOR field (HNSW algorithm) + TAG fields forapp_name/user_idscoped queriesFT.SEARCHwith KNN for vector similarity retrieval, pre-filtered by TAGadd_session_to_memoryandadd_events_to_memoryBatchpipelining for efficient ingestion (1 round trip regardless of event count)asyncio.Lockwith double-check locking for thread-safe index creationDependencies
valkey-glide >= 2.4.0(as optional dependency under[valkey]extra)valkey/valkey-bundleimage)Testing Plan
valkey/valkey-bundleimage) covering vector similarity search, user/app isolation, top-k limits, distance threshold filtering, metadata preservation, index idempotency