Use case · Personal AI

Two weeks later, your AI greets the user like a stranger

The job that changed. The goal that evolved. The hard week the user opened up about. Saved with a timestamp, read back at the start of the next chat.

The pain

The user shared a hard day. The assistant did not save the file.

A user opens the companion app on a Wednesday evening. They had a difficult conversation with their manager. They type into the chat, the assistant listens, the conversation winds down. The next morning, the assistant sees a clean slate. Two weeks later, the user types "I've been thinking about that conversation". The assistant has no record. The illusion of companionship breaks.

Built-in chat memory in most LLM platforms is shallow on purpose: a few recent turns, then a summary, then nothing. What a real companion needs is the opposite: a long timeline, with the ability to ask "how did this person feel about work three months ago" and get a coherent answer.

That answer cannot be a single summary string. It has to be a timeline. Preferences shift. Jobs change. Goals evolve. The assistant needs to know which version is current and which ones used to be true.

The shape of the fix

One memory per user, every fact stamped with a date

April — spring

New job, new context

  • User: "Just started at Globex"
  • Goal saved: ship weekly releases

Korely memory

Facts stamped with valid_at

  • One agent_id per user
  • Older facts marked superseded, not deleted

August — summer

Goal evolves, AI tracks the shift

  • User: "Weekly ship works, now reliability"
  • Old goal: superseded. New goal: current.

November — autumn

AI greets the user with full memory

  • "How is sleep going since June?"
  • "Are you still on the reliability push?"

The companion always knows the current state. The timeline is one search call away.

How Korely fits

One agent_id per user, timestamps on every fact

Your companion app calls korely.add() at the end of every meaningful exchange. The agent_id is the user's identifier; metadata anchors the fact to a point in time. The companion does not have to decide what to "remember forever" versus "forget" — the timeline handles it.

On the next chat, the companion reads the recent timeline and the relevant older facts before responding. "How did things go with the team release schedule?" answers itself, because the answer was saved in May and updated in August.

Privacy is a feature, not a footnote. Memory lives in our EU cloud — EU data residency by default, never the US. The user can export or erase everything stored about them with one call.

Show me the code

Save per session, search by topic or recency

personal_ai.py python 
from korely_memory import Korely

korely = Korely(api_key="kor_live_...")

# Wednesday evening — user shares
korely.add(
    "User had a difficult conversation with their manager. "
    "Emotional state: stressed. Cited reason: project deadline.",
    agent_id=f"companion-{user_id}",
    metadata={"timestamp": conversation.datetime.isoformat()},
)

# Two weeks later — user opens the app
recent = korely.search(
    "recent emotional events",
    agent_id=f"companion-{user_id}",
    limit=10,
)

# The companion opens with awareness, not amnesia
greeting = build_greeting_from(recent)

The companion writes once at the end of a meaningful exchange and reads once at the start of the next one. Everything else is application logic.

Frequently asked

Personal AI memory, common questions

Why do personal AI companions feel cold after a few weeks? +

Without long-term memory, each conversation is a fresh window. The companion treats every "what have you been up to" question as the first one. Real continuity requires persistent storage the assistant can re-read at the start of every chat.

How does Korely handle preference change without losing the old preference? +

Every memory carries a valid_at timestamp. When a preference changes, the new fact gets recorded and the older one is marked as superseded. Search returns the current state by default and the full timeline when asked.

Where does my personal AI memory live? +

In Korely cloud (EU) — a managed store your agent reads and writes over the API. Export or erase everything about yourself with one call. Want fully local notes instead? Korely also makes a desktop app, separate from the agent memory API.

How is this different from ChatGPT memory? +

ChatGPT memory is a black box managed by OpenAI. Korely stores typed facts you can read, export, or erase over the API, with an audit trail. The memory model is also richer: it tracks change over time, not just the latest fact.

Is it appropriate for sensitive personal conversations? +

You decide what gets stored. Korely does not coach the assistant on what to write; that is your application logic. For high-sensitivity contexts, EU data residency and one-call erase keep you in control.

Give your companion a memory worth growing into

The Python and Node SDKs are live. The Hobby tier is free, no credit card required.

Get started free

Looking for a different shape? See the other five use cases →