valkey.yaml

# Valkey Memory Store Configuration for Agentic Memory
# This configuration file contains settings for using Valkey (with the Search
# module) as the agentic memory backend.
#
# To use this configuration:
# 1. Set backend: "valkey" in global.stores.memory in your config.yaml
# 2. Inline the valkey settings from this file into global.stores.memory.valkey
# 3. Ensure Valkey server with the Search module is running and accessible
#
# Valkey Search module is required for FT.CREATE / FT.SEARCH vector operations.
# Use the valkey/valkey-bundle image or compile Valkey with --enable-search.

# Connection settings
host: "localhost"        # For production: use your Valkey cluster endpoint
port: 6379               # Standard Valkey port
database: 0              # Database number (0-15)
password: ""             # Authentication password (leave empty if not required)
timeout: 10              # Connection/request timeout in seconds

# Index and storage settings
collection_prefix: "mem:"   # Key prefix for HASH documents
index_name: "mem_idx"       # FT.CREATE index name
dimension: 384              # Embedding vector dimension (must match model)
metric_type: "COSINE"       # Distance metric: COSINE, L2, or IP

# HNSW index tuning
index_m: 16                 # Bi-directional links per node (higher = more accurate, more RAM)
index_ef_construction: 256  # Build-time search width (higher = better recall, slower build)

# TLS settings
tls_enabled: false                 # Enable TLS for the Valkey connection
tls_ca_path: ""                    # Path to PEM-encoded CA cert (empty = system trust store)
tls_insecure_skip_verify: false    # Skip server cert verification (development only)

# Full canonical config.yaml usage example:
#
# global:
#   stores:
#     memory:
#       enabled: true
#       backend: valkey                       # <-- select Valkey backend
#       auto_store: true
#       valkey:                               # <-- Valkey-specific settings
#         host: valkey
#         port: 6379
#         database: 0
#         timeout: 10
#         collection_prefix: "mem:"
#         index_name: mem_idx
#         dimension: 384
#         metric_type: COSINE
#         index_m: 16
#         index_ef_construction: 256
#         tls_enabled: false
#         tls_ca_path: ""
#         tls_insecure_skip_verify: false
#       embedding_model: bert
#       default_retrieval_limit: 5
#       default_similarity_threshold: 0.70
#       hybrid_search: true
#       hybrid_mode: rerank
#       adaptive_threshold: true
#       quality_scoring:
#         initial_strength_days: 30
#         prune_threshold: 0.15
#         max_memories_per_user: 200
#       reflection:
#         enabled: true
#         algorithm: recency_semantic
#         max_inject_tokens: 512
#         recency_decay_days: 14
#         dedup_threshold: 0.9
#
# Example configurations for different environments:
#
# Local Development (Docker):
#   host: "localhost"
#   port: 6379
#   password: ""
#
# Production (Docker / Kubernetes):
#   host: "valkey-service.valkey-system.svc.cluster.local"
#   port: 6379
#   password: "${VALKEY_PASSWORD}"     # from secret
#   index_m: 32                        # higher recall for production
#   index_ef_construction: 512
#
# Kubernetes with TLS:
#   host: "valkey-tls.valkey-system.svc.cluster.local"
#   port: 6380
#   password: "${VALKEY_PASSWORD}"
#   tls_enabled: true
#   tls_ca_path: "/etc/valkey/certs/ca.pem"   # mounted from secret