Long-Term Memory

With long-term memory enabled, an agent remembers facts and preferences about each person it talks to — across separate conversations. At the start of a call it recalls what it already knows, and during the call it can search its memory for specifics or save something new when asked ("remember that I prefer email").

Enabling memory

1. Open your agent in the Agent Builder.
2. In the settings drawer, expand the Memory section.
3. Check Long-term memory and save.
4. Publish the agent — memory applies from the next published version used in a session.

You can see whether memory is on for an agent at a glance without opening the builder: the agent's Overview page shows a Memory row (On / Off) next to its knowledge collections.

Memory options

Two settings sit below the Long-term memory switch. They fine-tune how memory behaves and only apply while memory is on:

• Reference previous conversations in the greeting (on by default) — when on, the agent may open a call by referring to something it remembers ("Welcome back — still using the address ending in …?"). Turn it off for a neutral greeting: the agent still knows the person's details and can use them naturally later in the call, it just won't lead with them.
• Memory capture — controls what new memories are allowed to form from:
  • Full conversation (default) — memories are extracted automatically from the whole conversation.
  • Only when explicitly asked — nothing is captured automatically. A memory is saved only when the person asks the agent to remember something ("remember that I prefer email"). In this mode the conversation transcript never leaves the call, so the agent can only remember what it was explicitly told to.

Both settings are versioned like every other builder change — they take effect from the next published version used in a session.

What the agent remembers

Memories are short, human-readable facts extracted automatically after each conversation (for example "Prefers email follow-ups" or "Has an open order A-1042"), plus anything the agent was explicitly asked to remember. Extraction runs in the background — a new memory becomes available from the next conversation, not within the same call.

Alongside individual facts, memory can also include a short summary of a past conversation and, where available, a compact session episode — a recap of what happened in a session and what the person was trying to do. These appear in the memory panels labelled as "Conversation summary" and "Session episode" so they're easy to tell apart from plain facts.

Memory is kept per person and per agent: what someone tells one agent is never visible to another agent, and one person's memories are never visible to anyone else.

Memory for portal members

Agents can also remember signed-in members of your portals — but only when you explicitly allow it. Member memory requires all three of these:

1. The agent has Long-term memory enabled (above).
2. The portal has Member memory enabled — open your portal, go to the Memory tab, and turn on "Let agents remember identified members between conversations".
3. The visitor is a signed-in member. Anonymous visitors are never remembered.

Because this stores personal data about your members, reflect it in your portal's terms or privacy notice when you enable it.

What members see

Members get their own Memories panel — reachable from the portal home page — showing everything the portal's assistants remember about them, labelled per assistant. There they can:

• Edit any memory inline (for example fix an out-of-date preference).
• Forget one memory — removes just that item.
• Forget everything — deletes all their memories along with the stored conversation history behind them. This cannot be undone.
• Turn remembering off — the "Let assistants remember me" switch stops new memories from being collected from their next conversation. Existing memories stay until the member deletes them; the member's switch always wins over your portal setting.

The panel always works, even after you turn Member memory off: the toggle controls whether new memories are collected, never a member's access to what is already stored.

Managing memories from the dashboard

You have two management views of your own:

• Agent → Memory tab — what the agent remembers about you from your own test sessions. Search it, forget single items, or clear everything for that agent.
• Portal → Memory tab — below the Member memory switch, an oversight list of what agents remember about each member, grouped per member. Use the per-member Erase action for support or data-deletion requests — it removes the member's memories together with the stored conversation history. You can erase member memories, but never edit them — editing stays with the member.

Retention

• Raw conversation transcripts — the recordings memories are learned from — are kept for 30 days, then deleted automatically.
• Extracted memories are kept until they're deleted — by the person from their own Memories panel, by you from the dashboard, or through a data-deletion request.

Deleting a memory also removes the stored conversation history behind it, and in Only when explicitly asked mode no transcript is ever stored in the first place.

Notes

• Memory is off by default for every agent, and Member memory is off by default for every portal.
• Turning memory off stops new memories from being collected; it does not delete existing ones.
• New and edited memories can take around half a minute to show up in lists — panels reflect your own changes immediately.