wingog

Inspiration

When working with multiple AI relay and proxy services, we found that each provider uses different model naming conventions for essentially the same underlying models. This makes switching providers, testing performance, or adjusting configurations unnecessarily time-consuming and error-prone. To solve this friction and improve development efficiency, we decided to build a unified aggregation tool that abstracts these differences away.

⸻

What it does

Wingog is an aggregation tool for AI relay services that provides a unified model mapping layer. It allows developers to use consistent model identifiers while seamlessly routing requests to different relay services, regardless of how each service names or structures its models. This makes switching providers fast, flexible, and configuration-friendly.

⸻

How we built it

We designed Wingog around a centralized model-mapping and configuration system that normalizes model names across different relay providers. The tool dynamically translates unified model identifiers into provider-specific names at runtime, allowing users to change or add relay services without modifying their application logic. The architecture focuses on simplicity, extensibility, and fast iteration.

⸻

Challenges we ran into

One major challenge was handling inconsistencies and frequent changes in model naming across different relay services. Some providers use similar names with subtle differences, while others rename models entirely, which required careful abstraction and flexible mapping rules. Balancing configurability with ease of use was another key challenge.

⸻

Accomplishments that we’re proud of • Built a flexible and extensible model-mapping system • Reduced provider switching time from minutes to seconds • Simplified multi-relay AI integration into a single workflow • Designed the project to be easily extendable for future models and providers

⸻

What we learned

We learned that abstraction is critical when dealing with fast-evolving AI ecosystems. Small inconsistencies, like naming differences, can significantly impact developer experience if not handled properly. We also gained deeper insight into designing systems that prioritize adaptability and long-term maintainability.

⸻

What’s next for wingog

Next, we plan to support more relay providers and introduce automatic model discovery and validation. We also aim to add better monitoring, logging, and performance comparison features to help users choose the best provider for their needs. Ultimately, our goal is to make multi-provider AI integration effortless and scalable.

Built With

• python