Skip to main content

Memory Management

Advanced techniques for managing memories in Vidurai.

Memory Lifecycle​

Store → Assign Importance → Decay Over Time → Compress/Evict

Importance Scores​

Automatic Scoring​

Vidurai automatically assigns importance based on:

  • Content semantic value
  • Position in conversation
  • Frequency of access

Manual Scoring​

# Critical information (never forgotten)
memory.store("Production API key: xyz...", importance=1.0)

# Important context
memory.store("User is debugging login flow", importance=0.8)

# Casual mention
memory.store("Weather is nice today", importance=0.3)

Decay Management​

When to Enable Decay​

✅ Chatbots with evolving conversations ✅ Long-running sessions ✅ Dynamic user preferences

When to Disable Decay​

❌ Knowledge bases with static facts ❌ Compliance/audit logs ❌ Critical business data

# Disable decay for permanent storage
memory = Vidurai(enable_decay=False)

Memory Inspection​

# Get all memories
all_memories = memory.get_all_memories()

# Check memory stats
stats = memory.get_memory_stats()
print(f"Total memories: {stats['total']}")
print(f"Token count: {stats['tokens']}")
print(f"Compression ratio: {stats['compression_ratio']}")

Memory Cleanup​

# Clear low-importance memories
memory.clear_below_threshold(importance=0.3)

# Clear by age
from datetime import datetime, timedelta
cutoff = datetime.now() - timedelta(days=30)
memory.clear_before_date(cutoff)

# Clear specific Kosha
memory.clear_kosha('annamaya')  # Clear working memory

Best Practices​

  1. Use appropriate importance scores

    • 0.9-1.0: Critical, permanent information
    • 0.7-0.8: Important context
    • 0.5-0.6: Normal conversation
    • 0.3-0.4: Casual mentions

  2. Monitor token usage

    if memory.get_token_count() > 50000:
        memory.trigger_compression()

  3. Periodic maintenance

    # Daily cleanup job
    memory.clear_below_threshold(0.2)
    memory.optimize_storage()

See Best Practices for production deployment tips.