Description
Every new AI conversation gets the same default title, "New AI Conversation". The title changes only if the user renames the conversation manually. In practice almost nobody does, so the recent-chats list and the history panel on the chat hub fill up with identical cards that can be told apart only by date.
Use case
A user starts several conversations during a day and later wants to return to one of them from the hub. With identical titles they have to open the conversations one by one. A title derived from what the conversation is about lets them find it at a glance.
Applications differ in how much they are willing to spend on this. Some want a good title and accept one extra short model call per conversation; others run on a strict token budget or on slow local models and prefer a zero-cost title; some rename conversations themselves or embed the chat with their own logic and want no automatic title at all.
Proposal
Introduce an application-level setting that selects how the title of a new conversation is produced, with three modes:
- First message. The title is the text of the first user message, trimmed to the title length limit. No model call, no delay. This is the default, so adding the add-on never causes additional model calls.
- Generated. After the first exchange (user message and assistant answer) the model is asked for a short title in a separate call. The call runs in the background and does not delay the assistant's answer; until it completes, the conversation shows the first-message title.
- Off. The static default title is kept, as today.
Description
Every new AI conversation gets the same default title, "New AI Conversation". The title changes only if the user renames the conversation manually. In practice almost nobody does, so the recent-chats list and the history panel on the chat hub fill up with identical cards that can be told apart only by date.
Use case
A user starts several conversations during a day and later wants to return to one of them from the hub. With identical titles they have to open the conversations one by one. A title derived from what the conversation is about lets them find it at a glance.
Applications differ in how much they are willing to spend on this. Some want a good title and accept one extra short model call per conversation; others run on a strict token budget or on slow local models and prefer a zero-cost title; some rename conversations themselves or embed the chat with their own logic and want no automatic title at all.
Proposal
Introduce an application-level setting that selects how the title of a new conversation is produced, with three modes: