ranade-oss makes provider choice less brittle

Users can now reliably enter their own credentials for the three core AI providers:

• Anthropic, for access to its Claude models.
• Gemini, Google's AI model family.
• OpenAI, including clearer recognition of the GPT-5.6 option.

The important detail is restraint. The change keeps OpenRouter, a service that routes requests across many model providers, limited to self-hosted deployments, while leaving approval controls for other providers in place. It also teaches model checking that GPT-5.6 may be listed under its underlying provider name, so an available model is no longer incorrectly shown as unavailable.

This is a small but meaningful reliability fix for teams evaluating which AI provider best fits their legal work.