Request: load personas directly from query string and offer "save" button

[buczek]

Currently, personas (=settings and conversation) are downloaded from an external URL referenced by parameter in the query string. It would be nice if a persona would also be accepted directly from a query parameter.

So additionally to 1 it should also accept something like 2 with the same result.

This would enable the web interface to offer a "Share" button which copies a generated URL of the current conversation state into the clipboard. See, for example "Share" - "Full Link" at https://godbolt.org/ where something like this is implemented.

One usage pattern would be to store the URL of a conversation as a bookmark or paste it into a notepad to resume the conversation from another browser.

Also it would make it more straight forward to publish a link to a configured conversation: You don't need to publish an additional file, just generate and publish a single link. Sharing of conversations with other users can also be done more ad-hoc and if you share the link by email, you don't have the privacy issue that you need the conversation data in a json file on a web server.

Further options:

• use some text compression scheme for the generated URL
• use some less ugly custom encoding
• accept the parameter from POST queries (possibly already done by the framework?)
• attach a link shortener service for content where there are no privacy concerns.

[buczek]

I just discovered, that that "Share" button more or less does much of this already. Model selection, prompt and settings are embedded in the url, just as I requested. The only delta to personas seems to be the that the dialog is not included. However, for many applications it should be enough to use what available via "Share".

This is not true for OpenAI models, however, because https://chat-ai.academiccloud.de, does not honor the system prompt for OpenAI models and the only way to prime it into something different would be via dialog. Also, if you want to give the Persona additional information you want to talk about, this should better go into the dialog.

But personally, I'm not too much interested in that. If my thoughts are correct, feel free to close the issue.

Addendum: However, Deepseek works best without a prompt but instructions in the dialog.