MirrorAPI and ToolBench key

[KyrinChen]

I think the the MirrorAPI server is built locally. do we need a extra key for usage permission?
In addition, I filled out the application form but received no response. I noticed that no one has got a key in the Toolbench issues for the past six months. maybe the Mirror API can be used as a substitute?

[importpandas]

Thanks for raising this! You can refer to the logic implemented in https://github.com/THUNLP-MT/StableToolBench/blob/aa4ed9f4737ad98bd706663f01d63623c3427812/server/main.py#L151. The server is designed to first try fetching real responses via RapidAPI, and fallback to LLM-based simulated responses only when the API call fails. So while it's possible to simulate all tool responses using LLMs (e.g., via MirrorAPI), this may introduce certain discrepancies or biases compared to the actual API results. For more details, you can also take a look at the modifications and optimizations they made to the server logic in the repo. This should give you a clearer idea of how they balance between real API calls and fallback responses.