Your AI Analyst Now Remembers - Introducing AI Memory

Until today, every conversation with the AI started from zero. You could tell it on Monday that your fiscal year starts in April, and on Tuesday it had no idea. Your colleague asking the same question in their own chat got an answer computed on the calendar year - technically fine, practically wrong.

Today we’re launching AI Memory. When you tell the AI something worth keeping, it saves the fact and references it in every future conversation on that property. Tell it once. It knows from then on - and so does everyone on your team.

---

How It Works

Memory is deliberately simple: a list of short, dated facts attached to each property.

When you state something durable in chat - “our fiscal year starts in April”, “the Brand campaign excludes competitor terms since July”, “our main KPI is newsletter signups, not pageviews” - the AI saves it. You’ll see a Memory updated chip in the conversation the moment it happens, so nothing is stored silently.

You can also be explicit: “remember that…”, “forget what I said about…”, or “actually, that changed - update it”. The AI updates or deletes the existing memory rather than piling up contradictions, and every memory carries the date it was last confirmed, so newer facts win.

From then on, every conversation on that property starts with the AI already knowing your saved facts. No tool lookup, no “as I mentioned last week” - it just knows.

---

Shared With Your Team, On Purpose

Memories belong to the property, not to the person who happened to mention them. When one teammate teaches the AI that staging traffic should be excluded, everyone’s analyses exclude it from then on.

This was a deliberate choice. Personal memory silos would mean two people asking the same question get different answers depending on what each of them individually told the AI - exactly the inconsistency an analytics team doesn’t want. One property, one memory, consistent results.

Every memory shows who saved it and when, so there’s no mystery about where a fact came from.

---

You Stay In Control

An AI that remembers things needs to be inspectable, and memory only works if you can trust what’s in it. So everything is visible and editable:

• The Memories page - every property has one, listing every saved fact. Edit anything that’s wrong, delete anything that’s stale, or clear the lot.
• Deletions apply immediately - a deleted memory is gone from the very next message, not “eventually”.
• Incognito chats - toggle incognito on any conversation and it neither reads nor writes memory. Useful when you want a fresh, unbiased take, or when you’re exploring a hypothetical you don’t want remembered.
• Per-property opt-out - switch memory off entirely in property settings if a property shouldn’t use it.

There are also guardrails on what gets saved in the first place. The AI only saves facts you state - never numbers from analytics data or content it fetched from tools - and it’s instructed to never save sensitive personal information.

---

Memory vs. Knowledge Base

If you’re using our Knowledge Base, you might wonder how the two relate. They’re complementary:

• Knowledge base is documentation you write deliberately - detailed markdown entries about data quirks, channel definitions, seasonal patterns. Think of it as briefing a new analyst.
• Memory is what accumulates naturally from working together - short facts captured in the flow of conversation, no editor involved.

The knowledge base is where you’d explain why traffic from India is excluded, with detail and history. Memory is where “fiscal year starts in April” lands the moment you happen to mention it. Both are searched and referenced by the AI automatically.

---

The Fine Print

Like everything else in AI Data Stream, memory follows our BYOK philosophy: the few hundred tokens of memory included in conversations go through your own API keys at cost, with no markup. Memories are stored in your team’s own isolated database, and deleting them really deletes them.

If your team has an OpenAI provider configured, memories are embedded for semantic search - the same mechanism the knowledge base uses - so the AI can find older memories by meaning once a property accumulates more than fit in a conversation. Without one, keyword search takes over. Either way, recent memories are always available.

Free plans include 20 memories per property; paid plans are unlimited. Since the most recent facts are always favoured, 20 well-chosen memories go a long way.

---

Try It

Open any property’s chat and tell the AI something it should know:

“Remember that our fiscal year starts in April.”

Then start a new conversation and ask when Q1 begins. That’s the whole feature - your analyst finally remembers what you tell it.

AI Memory is available now on all plans.