Rethinking How We Configure Models

I’ve spent a lot of time wrestling with configuration files that grow too fast. It usually starts simple, but then you add a second provider, a personal API key, and a work-related profile. Suddenly, your setup is a tangled web of environment variables and hardcoded strings that are hard to maintain.

Managing these models should be easier. I want to share some ideas we are exploring to make model configuration feel less like a chore. The goal is to move toward a system where you can switch between personal and work profiles without breaking your entire setup.

What You’ll Need

Before looking at these new directions, you should be familiar with our current implementation:

• Current Models
• Model failover
• OAuth + profiles

Quick Start

While these ideas are still in the exploration phase, I recommend following this path to keep your configuration clean. This approach focuses on predictable fallbacks and clear separation of concerns.

1. Use the provider/model format for all selections to keep naming simple.
2. Define a global fallback list so all sessions fail over in a consistent order.
3. Separate your text models from image-capable models to avoid routing errors.
4. Set up explicit auth profiles to distinguish between personal and work accounts.

Troubleshooting

Configuring multiple providers often leads to specific points of friction. Here are two areas we are currently investigating:

• Profile Rotation: We are deciding if rotation should happen at the provider level or the specific model level.
• Migration Paths: Moving from legacy configuration keys to a new structure requires a safe transition to avoid breaking existing sessions.

If you are unsure how to surface profile selection in your UI, it is best to stick to the explicit ordering defined in your provider settings for now.

Need help getting your current config running? Talk to the AI Setup Assistant.

What’s Next

• Learn more about Models
• Explore Model failover
• Set up OAuth + profiles