ranade-oss stops chat settings from fighting the lawyer
Jurisdiction and AI-model choices now survive the handoff from a first question into an ongoing legal conversation.
The fork fixes an irritating failure mode: a lawyer who selected Federal Canada or U.S. context could lose that choice as the first prompt opened the chat. That context now carries forward, including within the relevant project.
It also makes the model picker less brittle:
- Compatible models can be discovered from five already supported AI providers, rather than maintained only through a fixed list.
- Newly discovered models remain usable after selection instead of being rejected later.
- Reasoning controls appear only when the selected provider and model actually support them.
- Three explicit GPT-5.6 options join the existing general GPT-5.6 choice.
The result is a chat workspace that better reflects the jurisdiction and AI service a legal team deliberately chose.
So what Legal teams testing different AI providers or working across Canadian federal and U.S. contexts should care because basic chat configuration is becoming dependable enough to evaluate the underlying legal work.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?