Currently there are two different protocols for calling tools out in the wild: OpenAI's and Anthropic's. To support a broader ranger of LLMs it would be nice if FreeAIr supported both.
Here's the documentation of the protocols:
https://developers.openai.com/api/docs/guides/function-calling?api-mode=responses
https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview
Additionally the fact that only OpenAI's protocol is implemented leads to an almost endless re-calling when a tool fails due to a LLM supporting only Anthropic's protocoll. FreeAIr won't get anything useful out of an Anthropic-style response.
Implementing the protocol would close the gap to Claude.
Currently there are two different protocols for calling tools out in the wild: OpenAI's and Anthropic's. To support a broader ranger of LLMs it would be nice if FreeAIr supported both.
Here's the documentation of the protocols:
https://developers.openai.com/api/docs/guides/function-calling?api-mode=responses
https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview
Additionally the fact that only OpenAI's protocol is implemented leads to an almost endless re-calling when a tool fails due to a LLM supporting only Anthropic's protocoll. FreeAIr won't get anything useful out of an Anthropic-style response.
Implementing the protocol would close the gap to Claude.