IvanJS17 puts guardrails around bring-your-own AI
This fork makes each chat's AI choice a recorded decision, with safer handling when firm credentials change.
IvanJS17 turns a loose chat setting into a fixed route. When a chat is created, it is tied to a specific AI provider, model and version of the user's stored credential. Replace or revoke that credential and the old route stops working. It cannot quietly pick up whichever key is currently active.
The fork also adds:
- A route-selection screen when creating a chat, so users choose the provider and model up front.
- Live provider model lists, with a curated fallback if a provider cannot be reached.
- Support for several third-party AI services that use the same connection style as OpenAI.
That is a meaningful governance move, though it reaches across the chat data model and user interface.
So what Legal teams experimenting with multiple AI providers should care because it creates a clearer audit trail and avoids silent credential changes.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?