Short-term Memory

When to Use

Use short-term memory as a "scratchpad" or RAM for the current session. Data here is temporary and meant to expire. Perfect for:

• Current conversation topic
• User's immediate mood or sentiment
• Drafting responses or holding intermediate reasoning steps

Key Operations

Code Example

Here we buffer the user's current mood and topic. Note the ttl_seconds - this data will auto-delete after 1 hour.

# 1. Add to Short-term Buffer
session_id = "session_week1_day4"

# Store current context with 1 hour TTL
client.memory.short_term.add(
    session_id=session_id,
    key="current_topic",
    value={"topic": "running", "subtopic": "knee_pain"},
    ttl_seconds=3600
)

client.memory.short_term.add(
    session_id=session_id,
    key="user_mood",
    value="anxious",
    ttl_seconds=3600
)

# 2. Retrieve Context during conversation
topic = client.memory.short_term.get(session_id=session_id, key="current_topic")
print(f"Agent should focus on: {topic}")

Key Takeaways

• Short-term memory is key-value based, making it fast and direct.
• TTL (Time To Live) ensures you don't clutter storage with stale context like "user is typing...".
• It complements Semantic/Episodic memory by handling "right now" vs "forever".