Memory
What the Agent remembers about you and your organization, where to see, edit and delete it, and what it is told never to keep.
Availability: Available on every plan. Suggestions from conversations spend AI credits. Compare plans
Who can do this: Every member, for their own memory. Owners and Admins, or a role with Organization settings, for the organization's.
The Agent keeps short statements and carries them into your conversations. A fact says what is true, such as "Our fiscal year starts on 1 April". A rule says how to work, such as "Quote every amount with its currency".
Two memories#
| Your memory | Organization memory | |
|---|---|---|
| Where | Settings > Account > Your memory | Settings > Agent and AI > Agent, under Organization memory |
| Who it reaches | Your own conversations only | Every member's conversations |
| Who decides | You | Owners and Admins, or a role with Organization settings |
| What fits | How you like answers, how you sign off, what you always want done | What holds for the whole organization |
In a conversation, the Memory row in the composer's + menu lists what the Agent carries into it, each marked Whole organization or Only you. It links to Manage memory and to Your memory.
Add a statement#
Open the memory#
For yours, open Settings > Account > Your memory. For the organization's, open Settings > Agent and AI > Agent and find Organization memory. Choose Open memory, or Review all N when suggestions wait.
Write it#
Choose Add a fact or rule. Under "One fact or rule the Agent should always know", write one statement of 3 to 500 characters, choose Fact or Rule, and choose Remember.
You can also ask in a conversation, for example "Remember that our fiscal year starts on 1 April". The Agent's save is a card in every autonomy mode, so nothing is kept until you approve it. A member's note for the organization then waits for an admin to keep it.
Waiting, Kept and Dropped#
Each memory sorts its statements into three tabs. Only what is under Kept reaches the Agent.
| Tab | What is there | What you can do |
|---|---|---|
| Waiting | Suggestions nobody has decided on | Keep or Drop. Select several for Keep N or Drop N. |
| Kept | What the Agent carries into conversations | Edit, then Save. Or Drop. |
| Dropped | What was dropped | Restore it. It comes back in effect. |
A row says where it came from: Suggested by the Agent from N conversations, Suggested by a member or Added by hand. A suggestion that replaces a kept statement shows Replaces: with the old words struck through, and keeping it takes the old statement out of use. Search memory and the kind filter (Facts, Rules) narrow a long list.
In the organization's memory only an admin keeps, drops, edits or restores. No agent, workflow or API key can put a statement in effect.
Suggestions from conversations#
The Agent can suggest statements from your conversations. It is off until someone turns it on, because it spends AI credits.
- Organization. An admin turns on Learn from conversations under Organization memory. For anyone else the switch is locked and reads "An admin decides". Nothing counts until an admin keeps it.
- You. Turn on Learn from my conversations on Your memory. Nothing is kept until you approve it.
Once a day the Agent reads up to 20 of the newest conversations since its last read and suggests at most 5 statements. Each waits under Waiting. It skips the day when AI is off, when credits are out or when nothing is new.
The organization's memory learns from members' conversations only. A member who turns off Let the organization's memory learn from my conversations, under Sharing on Your memory, is left out. It is on by default.
What it is told not to keep#
When it suggests statements, the Agent is told to leave out:
- For the organization. Personal trivia, one-off tasks, secrets, credentials and anything about a single record.
- For you. Anything about the organization, a customer, a colleague or a record, and one-off tasks, secrets and credentials. Your memory holds how you like answers, how you sign off and what you always want done.
A suggestion always waits under Waiting until a person keeps it. Your memory is yours alone: no agent, workflow or organization key can read or change it, and neither can an admin.
Erase your memory#
On Your memory, Erase your memory deletes everything the Agent keeps about you. Choose Erase, then Erase for good. It cannot be undone, and it turns learning off.
This is the one delete. Dropping a statement, in either memory, keeps it under Dropped.
Questions#
Does a rule change what the Agent may do? No. A rule is words the Agent reads. What it may do is set by the person's rights and the autonomy mode. See Autonomy and approvals.
Does memory spend AI credits? Suggestions from conversations do, as the switch says.
Where does the Agent keep how to do a job? In a skill, not in memory. See Skills.