User Profiles

One call that answers "who is this user, currently?" — built from the collection's latest, non-expired facts and preferences.

Get a profile#

# Python
p = mem.profile("people")
print(p["profile"], "based on", p["based_on"], "memories")

# HTTP
GET /v1/memory/collections/people/profile
→ {"profile": "Alex is a product manager at Stripe who prefers morning meetings…",
   "based_on": 12}

What goes in#

• Only fact and preference memories — episodes are excluded.
• Only current ones: superseded (is_latest = false) and expired facts never appear, so the profile tracks reality as it changes.
• Ranked by importance and recency (up to 80 memories feed the summary).

Using it in an agent#

profile = mem.profile(f"user-{user_id}")["profile"]

system_prompt = f"""You are a helpful assistant.
What you know about this user:
{profile}
"""

Pattern: one collection per end-user (or one collection + container tags), remember what conversations reveal, and fetch the profile at session start. Profiles are computed on demand by an LLM call — cache per session, not per message.