Understanding Agent Memory

Agent Memory is where the AI Agent keeps what it has learned about your organisation. Every time you tell it something durable in a conversation, for example your value proposition, who you target, or how you like sequences written, it can store that as a memory and reuse it later. The result is that you stop re-explaining your business every time you open a chat.

You will find it in Settings > Agent Memory.

What you see on the page

The page lists every memory the agent holds, with four columns:

Use Search, Filters and Columns at the top to narrow the list down when it grows.

Where memories come from

There are two routes.

The agent adds them itself

During a conversation the agent extracts what looks important and stores it. You can also make this explicit: tell it "store this as a memory" and it will.

You add them manually

Click + Add memory on the Agent Memory page and fill in the form:

1. Category (required). Pick from Value Proposition, Target Audience, Preferences or Sequence Learnings.

2. Kind (required). Durable memories persist long-term. Episodic memories capture time-bound learnings and expire after 90 days.

3. Content (required). The fact itself. Write it as a standalone statement, since the agent reads it without the conversation it came from.

4. Importance. A slider from 1 to 10, default 5. Higher importance memories are prioritised in agent conversations.

5. Scope. Off means personal, so only you see it. Toggle Organization on to share it with all members.

Click Create and it is live for the next conversation.

When the agent adds a memory itself, it does so a little while after the conversation goes quiet, not instantly. If you want something recorded on the spot, say "remember this".

There is no cap on how many memories you can store.

Agent Memory or Organization Context?

Both feed the agent, and they are not meant to overlap.

Organization Context holds the foundation: what your company does, what you sell, who you sell to. See 📄 Organization Context 101.

Agent Memory holds everything that is not in there. Your outbound routines, the tone you write in, your preferences, what you have learned from past sequences. You do not need to repeat anything already stored in Organization Context.

If you say "remember this" about something that is actually a fact about your business, your ICP, value proposition or positioning, the agent redirects it into Organization Context instead, with your approval, rather than saving it as a memory. So if you expected a memory and Agent Memory looks empty, check Organization Context first.

Choosing Durable or Episodic

• Durable for things that are true until you change your positioning: what you sell, who you sell to, how you want messages written.

• Episodic for things tied to a moment: a campaign that is running now, a test you are learning from this quarter, a temporary priority. These clear themselves out after 90 days, so the agent is not working from stale context a year later.

Choose carefully, because Kind cannot be changed once the memory is saved. An Episodic memory cannot be promoted to Durable, or the other way round. If you got it wrong, delete the memory and create it again.

Editing and deleting a memory

Scroll to the right of the memories table, click the three dots in the Actions column, then choose Delete.

Anyone in your organisation can edit or delete an Organization-scoped memory. Treat them as shared team settings rather than personal notes.

Writing a good memory

• One fact per memory. Split anything with an "and" in it.

• Be specific. "We target Heads of Sales at 50 to 500 employee software companies in Western Europe" beats "we target sales leaders".

• Reserve high importance (8 to 10) for the handful of facts that should shape every answer. If everything is important, nothing is.

• Use Organization scope for anything true of the company, and personal scope for your own working preferences.

Still have a question? See 📄 FAQ: AI Agent, Memory, Skills & Playbooks.