​

Memory Configuration

​

Storage Configuration

​

In-Memory Storage

memory:
  type: "in_memory"

• No configuration required
• Automatically enabled by default
• Fast performance
• No persistence (data lost on restart)

• Testing and development
• Temporary tasks
• Quick prototyping
• Low-memory requirements

​

Vector Database Storage (LanceDB)

memory:
  type: "lancedb"
  db_dir: "/path/to/memory/database"
  collection_name: "memories"
  embedding_model: "text-embedding-v4"  # Optional
  similarity_threshold: 0.8

• Directory path for LanceDB storage
• Automatically created if doesn’t exist
• Persists across restarts

• Name of the LanceDB collection
• Multiple agents can share or use separate collections
• Useful for organizing memory by agent or user

• Embedding model for semantic search
• Uses configured embedding models if not specified
• Required for semantic similarity search

• Minimum similarity score for memory retrieval
• Range: 0.0 to 1.0
• Higher = more strict filtering
• Lower = more memories retrieved

​

Search Configuration

​

Similarity Threshold

• Only very similar memories retrieved
• Fewer results, higher precision
• Best for: Specific, focused domains

• Balanced similarity and coverage
• Recommended for most use cases
• Good mix of precision and recall

• More memories retrieved
• Higher recall, lower precision
• Best for: Exploratory tasks, diverse patterns

​

Result Limit (k)

memory:
  search:
    k: 5  # Return top 5 most relevant memories

• Small (3-5): Focused, fast retrieval
• Medium (5-10): Balanced performance
• Large (10+): Comprehensive context, slower

​

Search Types

• Uses vector similarity
• Finds conceptually related memories
• Best for: Pattern matching, analogies

• Keyword matching
• Exact phrase matching
• Best for: Specific terms, precise lookups

​

Memory Management

​

Automatic Cleanup

memory:
  cleanup:
    enabled: true
    retention_days: 90
    min_access_count: 2
    deduplicate: true

• Delete memories older than specified days
• Default: No automatic deletion
• Set based on your needs (e.g., 30, 90, 365)

• Keep frequently accessed memories
• Delete unused memories below threshold
• Helps maintain quality

• Automatically merge duplicate memories
• Combines similar content
• Reduces memory size

​

Manual Management

• View memories - Browse stored memories
• Delete memories - Remove specific memories
• Export memories - Download memory data
• Clear all - Delete all memories (reset)

​

Performance Tuning

​

For Speed

• Use in-memory storage
• Lower k value (fewer results)
• Higher similarity threshold
• Filter by category

memory:
  type: "in_memory"
  search:
    k: 3
    similarity_threshold: 0.8
    filters:
      category: "execution"

​

For Quality

• Use LanceDB with semantic search
• Medium k value (5-10)
• Medium similarity threshold
• No aggressive filtering

memory:
  type: "lancedb"
  db_dir: "/var/lib/xagent/memory"
  embedding_model: "text-embedding-v4"
  search:
    k: 8
    similarity_threshold: 0.6

​

For Storage Efficiency

• Enable automatic cleanup
• Short retention period
• Aggressive deduplication
• Periodic export and archive

memory:
  type: "lancedb"
  cleanup:
    enabled: true
    retention_days: 30
    min_access_count: 3
    deduplicate: true

​

Agent Configuration

• When to store memories
• What to store in each memory
• Which memories to retrieve for each task
• How to use retrieved memories

​

User Privacy Options

• Check if memory is enabled
• View stored memories
• See what Xagent has learned

• Delete memories - Remove specific or all memories
• Export data - Download your memory data
• Disable memory - Contact admin to disable memory for your account

​

Troubleshooting

​

Memory Not Working

• Memory is enabled in configuration
• Storage backend is accessible
• Embedding model configured (for LanceDB)
• Sufficient disk space

​

Poor Search Results

• Adjust similarity threshold
• Increase k value
• Check memory content quality
• Verify embeddings are working
• Try text search fallback

​

Slow Performance

• Reduce k value
• Increase similarity threshold
• Filter by category
• Consider in-memory storage
• Check database performance

​

High Memory Usage

• Enable automatic cleanup
• Shorten retention period
• Increase min_access_count
• Deduplicate more aggressively
• Export and archive old memories

​

System Configuration Examples

​

For Development Environments

memory:
  type: "in_memory"  # Fast, no persistence needed

• Fast performance for testing
• No data persistence across restarts
• Suitable for development and testing

​

For Production Environments

memory:
  type: "lancedb"
  db_dir: "/var/lib/xagent/memory"
  embedding_model: "text-embedding-v4"
  search:
    k: 5
    similarity_threshold: 0.7
  cleanup:
    enabled: true
    retention_days: 90
    deduplicate: true

• Persistent storage
• Semantic search capabilities
• Automatic cleanup and maintenance
• Suitable for production use

​

For Privacy-Sensitive Applications

memory:
  type: "lancedb"
  db_dir: "/secure/memory"
  encryption: true
  cleanup:
    enabled: true
    retention_days: 30

• Encrypted storage
• Shorter retention period
• Enhanced privacy protections

​

Monitoring

​

Memory Statistics

• Total memories stored
• Retrieval frequency
• Average relevance scores
• Category distribution
• Storage size

​

Performance Metrics

• Search latency
• Retrieval accuracy
• Memory hit rate
• Deduplication savings

​

Next Steps

• Memory Overview - Learn about memory concepts
• Memory Types - Understand memory categories
• Building Agents - Create memory-enabled agents