Would you consider having the contents of the Chat Inspector be unique to each record?
It woud be really helpful when selecting a Group or Document if the Chat Inspector contained the most recent chat session about that record (unless cleared by the user).
The most recent chats are accessible via the popup in the inspector and restore the used document/selection. Frequently chats are not related to any documents at all or to multiple ones or the selection changes during the session. There could be even multiple chats related to the same document.
Would it be possible to allow the user to rename chats? That would be very helpful to be able to find them in the future
Also would it be possible to pin specific chats so something particularly important stays at the top and is not archived off?
Perhaps let the user decide how many chats are saved?
See Settings > AI > Chat. But recent chats are just convenience, just like any recent list. I would recommend saving chats instead.
Thanks
The ability to pin so a particular chat does not roll off would be helpful.
Yes you can save chats - but there are times when it is helpful to be able to save it in a way that I can go back and continue the chat in the same window. Right now I can only do that until the chat rolls off as the oldest one. I might have some chats that I want to keep indefinitely and perioidically return to
BTW:
Starting a fresh chat as often as possible is widely considered a good practice. This does not only save time, money & tokens but also improves the responses, e.g. by avoiding confusing or contradictory statements or past errors in a longer conversation (“context rot”). In addition, modern LLMs might handle a context window of up to 1M tokens but the quality and accuracy degrades the longer the chat gets and initial instructions or roles might get lost or be ignored.
Frequent fresh chats certainly were the initial recommendation when AI was evolving and I agree that still makes sense in many situations.
But that seems to be evolving. Claude Dispatch for example is explicitly one continuous thread in order to maintain context. (It automatically does compacting periodically when it is approaching max context window size.)
Example of how an ongoing chat can be helpful.. Suppose I am planning a flight as a small airplane pilot and I ask Claude to retrieve weather forecasting data starting a week in advance - and there are concerns about an evolving major weather system. Rather than repeating the entire analysis many times it is helpful to simply prompt “Get current weather data and advise of any significant changes.”
Another example - Suppose there is a topic in medicine or any other academic or professional area which is my particular focus. Rather than running a complete literature search weekly if I keep a Chat open I could simply prompt “Search for any updated publications.”
Another example would be if I am following a particular legal case and I want to do a daily search for updates in the legal docket. Now that Claude Code is configured as my custom OpenAI LLM, I can create a Devonthink Skill to search the legal docket via an external MCP server. It would be very helpful to keep that as an ongoing chat so periodically search for any updates.
It has not stopped evolving and the AI companies still advocate what Criss mentioned.
And I think either approach can make sense depending on the situation.
So why not let the user decide?
It‘s noted but as usual no promises.