Memory and learning: Concierge docs

Concierge agents keep short memories about how your workspace works: "We ship replacements for damaged items without waiting for a return", "Active subscriber means an active contract, not paused." They use them when they plan and reply, so your team stops repeating itself. Memories are about how you work, never about a specific customer. Find them under Build > Memory.

Where memories come from

Telling an agent. Say "From now on…" in a chat with an agent, for example "From now on, put refunds first in the weekly report." The agent saves it right away and shows "Saved to memory" under your message, with Undo and Edit. Saying the same thing again confirms the existing memory instead of adding a copy.
Handoffs. After your team handles a conversation an agent handed off, the agents suggest what they learned: an answer your team gave, how the team resolves that kind of request, or what the team needs from the customer. In the inbox you'll see "Your Agents learned … from how this was handled · Review".
Work that went well. After a job that used tools succeeds, the agent writes a short journal of the steps that worked and uses it on similar work.
You write it. Click + New memory on the Memory page.

Who a memory is used with

Every memory has one audience. Use the View menu on the Memory page to switch between them:

Customer-facing: used in conversations with customers and with teammates. Handoff lessons are always customer-facing.
Team: used only in conversations with teammates, for preferences and corrections like report formats.
Only me: used only in your own private conversations with your agents. Only you see them.

The View menu also has Journals and, for teammates who manage Concierge, Suggested.

Write a memory

New memory

On Build > Memory, click + New memory.

Title and type

Give it a short Title (up to 80 characters, for example "Damaged items") and a type: Preference, Correction, How the team works, What customers can expect, Definition of a term or Answer the team gave.

The memory

Write it in up to 600 characters, as one clear rule or fact.

Who it's used with

Choose Customer-facing, Team or Only me.

Confirm and save

Tick "I confirm this names no customer and holds no private information, such as names, emails, phone numbers, addresses or order details." and save.

Review suggestions

Open View > Suggested. Each suggestion shows what it says and where it was learned. Click Approve, Edit & approve or Discard. When a suggestion would replace an existing memory, you see both ("Now" and "Suggested"); approving it archives the old one.

By default, an answer your team gave goes live right away, and other suggestions go live once handoffs from two different customers back them up. To approve everything yourself, open Learning settings and set Customer-facing suggestions to "Use a suggestion only after someone approves it".

Learning settings

Click Learning settings on the Memory page. All three switches are on by default:

Learn from handoffs: suggest lessons after your team handles a handed-off conversation.
Keep journals: note the approach that worked after successful work with tools.
Learn from teammates: remember it when a teammate says how things should be done from now on.

What's kept out

Learning keeps practices, not personal details. Before a model reads a handoff, the customer, their orders and the teammates involved are replaced with placeholders. Anything learned automatically is checked again and refused if it contains an email address, phone number, order number, tracking number or a link with a token, or if it describes a specific person, order or case. If you tell an agent something about one customer, you'll see "Not saved to memory: it names a customer. Save it as a note on that customer instead."

How agents use memories

For each request, the agent picks the few memories and journals that fit it best. Pinned memories ("Pin: always use") are always included.
Customer conversations use only customer-facing memories. Teammate conversations add team memories, and your own memories in your private conversations.
A memory nobody has confirmed in 180 days shows "Not in use" and is skipped. Editing or pinning it puts it back in use, and every time a successful job applies a memory, it counts as confirmed.

Archive, forget, delete and export

Archive stops agents using a memory. Restore brings it back.
Forget erases a memory's text for good. It can't be restored, and no agent, page or export can use it again. You can also ask an agent in chat to forget one of your own or a team memory; customer-facing memories are forgotten on the Memory page.
Delete removes an archived memory permanently.
Archive all, in Learning settings, archives every customer-facing and team memory, including suggestions. Teammates' own memories stay.
On View > Only me, Export my memories downloads your personal memories as a file, and Forget all my memories erases them all.

Memory or playbook?

A memory is one rule or fact, used wherever it's relevant. A playbook is a whole method for one kind of job. On the Journals view, Turn into playbook starts a draft playbook from an approach that worked.

Troubleshooting and FAQ

The agent isn't using a memory

Check that its status is In use and that its audience fits the conversation. A Team memory is never used with customers. If it should apply to every request, pin it.

"Some pinned memories are left out"

Agents use a limited number of memories per request. Unpin the ones that don't apply to every request.

Who can change memories?

Anyone with Concierge access can see them and manage their own. Creating, approving, deleting and Archive all need permission to manage Concierge conversations.

I saved something by mistake in chat

Click Undo under your message, or forget it on the Memory page.

Guides

On the product page

Learn more

What is AI memory?

Memory and learning