Examples

Pruning Engine

Manage memory lifecycle with intelligent forgetting policies.

Lifecycle Module

Scenario: Forgetting old weather reports

When to Use

Infinite memory is expensive and noisy. Use the Pruning Engine to automatically remove low-value information based on policies. Perfect for:

Removing old "small talk" (weather, greetings)
Enforcing data retention policies (e.g., "Delete financial data after 7 years")
Cleaning up temporary facts

Key Operations

set_ttl()

Define retention policy for item types

analyze()

Dry-run to see what would be deleted

Code Example

We set a policy that "weather reports" should be forgotten after 1 day, while "medical history" is kept for a year.

# 1. Set Retention Policies (TTL)
client.memory.pruning.set_ttl(
    item_type="weather_report", 
    ttl_seconds=86400, # 1 day
    priority=0 # Low priority
)

client.memory.pruning.set_ttl(
    item_type="family_medical_history", 
    ttl_seconds=31536000, # 1 year
    priority=10 # High priority
)

# 2. Run Pruning Analysis (Dry Run)
report = client.memory.pruning.analyze(
    kg_name="KG_Companion_Demo",
    criteria={"expired_only": True},
    dry_run=True
)

print(f"Candidates for deletion: {report['candidates_count']}")

Key Takeaways

Pruning prevents context pollution. Too much irrelevant memory confuses the LLM.
Structure your data with item_type (e.g., 'weather', 'medical') to enable granular pruning policies.
Always run analyze() (dry run) before execute() in production.