Memory System¶

FlowAgents provides long-term memory capabilities powered by mem0.

Overview¶

Memory enables agents to remember user preferences and information across conversations. When enabled, the system automatically:

Recalls relevant memories before agent execution
Stores collected information after successful completion

Configuration¶

YAML Configuration¶

# config.yaml
memory:
  enabled: true
  use_platform: true
  api_key: ${MEM0_API_KEY}

Self-Hosted mem0¶

memory:
  enabled: true
  use_platform: false
  vector_store_provider: qdrant
  vector_store_config:
    host: localhost
    port: 6333
  llm_provider: openai
  llm_model: gpt-4o-mini
  embedder_provider: openai
  embedder_model: text-embedding-3-small

Per-Agent Configuration¶

Enable memory for specific agents in workflow:

# workflow.yaml
agents:
  FlightAgent:
    enable_memory: true
    memory_config:
      recall_limit: 10
      store_on_complete: true

  HotelAgent:
    enable_memory: true

Field Filtering¶

Control which fields are remembered:

memory:
  enabled: true
  api_key: ${MEM0_API_KEY}
  remember_fields:
    - email
    - phone
    - preferences
  exclude_fields:
    - password
    - credit_card

Auto Behaviors¶

memory:
  enabled: true
  api_key: ${MEM0_API_KEY}
  auto_recall: true   # Recall memories before agent runs
  auto_store: true    # Store collected fields after completion

Best Practices¶

Use mem0 platform for production - Managed infrastructure, no setup
Filter sensitive fields - Use exclude_fields for passwords, tokens
Set field allowlist - Use remember_fields to limit what's stored
Test with self-hosted first - Use local Qdrant for development