“For example, a game developer might use LLM to generate game character descriptions or parameters on the fly. To make this work, LLM needs to reliably ‘generate’ that information in valid JSON format. Currently, developers have to rely on third-party tools like Pydantic, Zod, LangChain, etc. to repeatedly run the same prompts until inference returns usable data, which is not a good solution,” Shimmin said.
Analysts cited Phi-based model NuExtract as an example, explaining that what developers want is a model that knows how to properly use APIs and formatting languages such as JSON, with the ability to call functions within the model that have that functionality integrated into the model syntax.
Why is Salesforce releasing this product as ‘open source’?
Analysts believe Salesforce is releasing these products as “open source” to gain market share for features like Agentic tasks.
“Salesforce aims to make this tool available to developers and researchers to accelerate the development and improvement of function call models, potentially leading to more robust and reliable AI applications,” Hinchcliffe said, adding that open sourcing APIGen could lead to more specialized and efficient AI solutions that better suit the unique needs of different industries, improving business operations and customer experiences.
