Memory for customer support agents

What you are building

An AI agent (or copilot for human agents) that answers tickets and chats across email, chat and phone, for many customer accounts.

Why memory decides whether it works

• Customers repeat themselves when context is lost between channels and shifts.
• Account facts change — plan, contacts, open incidents — and a stale fact produces a wrong answer that sounds confident.
• One customer’s data must never appear in another customer’s conversation.

What Auld provides

• Per-account audiences so recall is scoped to the customer and your team.
• Supersession for plan, contact and status changes, with history kept.
• Sources on every memory so agents can say where a fact came from.
• Webhook sources for helpdesk and CRM events (Flex and Enterprise).

Getting there

1. 01

   Map accounts to audiences

   One audience per customer org; add its users and your support staff.

2. 02

   Ingest ticket events

   Comments, status changes and resolutions as records with real times.

3. 03

   Recall per conversation

   Before each reply, recall with the customer and the ticket subject.

4. 04

   Show sources to humans

   In copilot mode, display the cited source next to each suggested fact.

Data and boundaries

Scenarios on this page are illustrative and do not describe a specific customer.

Frequently asked questions