plugins must be deactivated and reactivated

[ninodamico]

Good evening, it happens very frequently that while you are chatting with the cat and you have an active plugin, if the cat, after calling the tool (internal to the plugin) at the first question, is unable to do so at the second question, the plugin must be deactivated and reactivated otherwise it is practically dead for subsequent questions. If you send a chatbot into production it's a problem. How can I solve it?

[pieroit]

Moving this issue to core repo as it is not related to local-cat.

Can you track doewn the error in the terminal?
When you report a bug, always tell how to replicate it, otherwise it is impossible to help

[valentimarco]

Moving this issue to core repo as it is not related to local-cat.

Can you track doewn the error in the terminal? When you report a bug, always tell how to replicate it, otherwise it is impossible to help

i don't think is related to local-cat!

Like @pieroit said, we need an dump of the terminal in order to understand the possible bug.

[ninodamico]

I'll show you in the screenshot, it's the behavior of the AiChatSql plugin, if in one question it stops querying the db, then in the next one it's as if it has been deactivated. So to make it work again it must be deactivated and reactivated manually.

Registrazione.2024-11-12.201519.mp4

[valentimarco]

I'll show you in the screenshot, it's the behavior of the AiChatSql plugin, if in one question it stops querying the db, then in the next one it's as if it has been deactivated. So to make it work again it must be deactivated and reactivated manually.
Registrazione.2024-11-12.201519.mp4

Seems a broken plugin... Can you try to create a similar plugin or even a different one and try again?

[ninodamico]

I can intervene on the plugin code, try to modify it, I wanted to intervene with a hook, that if the plugin stops being used, it deactivates and reactivates the plugin, can this be done?! Another problem is when I want to interrupt the flow with a response in case the tool is not invoked, so the LLM does not formulate response taken out of the db, but I can't use a control variable(
is the boolean variable called triggered in the screen). I'll show you in the screen, if you know how I can do it !!! also I was thinking that for both problems using an attribute that contains the name of the tool inside a hook (for example: in agent_fast_replay ) could be a solution. I should locate it in the source code, but where?