Long-term Memory

When to Use

Use long-term memory for "consolidation". Only high-value episodes or summaries should move here after a session ends. Perfect for:

• Summaries of completed projects or conversations
• Critical life events (e.g., "Graduated college", "Moved city")
• Core memories that define the user's history

Key Operations

Code Example

After a stressful week where Emma's grandmother was hospitalized, we archive the summary of this event so the agent remembers it forever.

# 1. Archive a consolidated memory
client.memory.long_term.archive(
    episode_id="ep_session_week2_day5",
    kg_name="KG_Companion_Demo",
    data={
        "summary": "Nana Rose hospitalized with heart palpitations, Emma very stressed.",
        "outcome": "Stabilized",
        "impact": "High"
    },
    user_id="emma_2024",
    metadata={"importance": "critical"}
)

# 2. Retrieve years later
# "What happened with Emma's grandmother?"
results = client.memory.long_term.retrieve(
    query="grandmother hospitalization history",
    user_id="emma_2024"
)

Key Takeaways

• Long-term memory is for archival. Think of it as the "File Cabinet" vs Episodic's "Diary".
• Items here are meant to be permanent and highly searchable.
• This module supports versioning, so you can update a long-term memory as new details emerge.