Conversations and context

A conversation is a chat between a customer, Tori, and any members of your team who join in. Tori uses the messages in that conversation to understand what the customer needs.

Each conversation has its own history. Sources and custom instructions are shared across conversations with an agent, but the messages are not.

What Tori knows in a conversation

Tori can use what the customer has already said, its own earlier replies, and messages your team has sent to the customer. It can look back through older messages when it needs more context.

For example, if the customer has already explained which product version they use, Tori can use that information when answering a follow-up question. The customer doesn't need to include it in every message.

Tori doesn't learn facts that it can carry from one conversation into another. Even when the same customer returns, Tori doesn't use their other conversations to answer a new one.

Continuing or starting again

Customers can continue a conversation for as long as they want. They can also start a new conversation and return to an old one later.

The choice is about whether the earlier discussion is useful for the next question. Continuing a conversation lets Tori use what has already been discussed. Starting a new one gives it a fresh conversation history.

Starting a new conversation does not remove the old one. It also doesn't change the agent's sources or instructions.

An existing conversation uses the agent's current configuration when Tori responds. It is not a saved copy of the agent as it was when the conversation began.

Returning to a conversation

The chat remembers anonymous visitors in their browser so they can return to earlier conversations. This doesn't require the visitor to create an account.

If they clear their browser data or use a different browser or device, the chat may no longer recognise them as the same visitor. This can prevent them from finding their earlier conversations.

You can also identify logged-in customers using accounts from your own website. This lets the chat show the same customer's conversations when they sign in on another device.

Customer identity and account information

Identifying a customer tells the chat which conversations belong to them. It doesn't give Tori access to their account in your product.

Tori doesn't automatically know the customer's plan, orders, account settings, or the page they are viewing. It can use information they provide in the conversation, but it can't look up those details in your systems.

Messages and responses

Tori can send several messages while responding to one customer message. For example, it can explain part of an answer, do more research, and then send the rest.

If Tori asks a question and waits for the customer to reply, its answer after that reply is a new response. The clarifying question and the later answer are two separate responses.

Corrections and testing

A correction in a conversation becomes part of that conversation's history. Tori can use it in later replies in the same chat.

That doesn't make the correction available in other conversations. To change the information Tori uses across conversations, update a source or custom instruction.

This matters when testing. If you tell Tori the correct answer and then ask the same question again, you are testing whether it can use what you just said. To check whether your sources and instructions are enough, ask in a fresh conversation without giving it the answer first.

Messages from your team

Messages your team sends in the customer conversation are also available to Tori. By default, Tori treats those messages as company information and can follow instructions your team gives it for that conversation.

Your team's separate internal discussion is not part of the history Tori reads. See Working with humans for how customer replies and internal notes fit together.