functionary

Why we included this project

Functionary is a good reference if you want to see how a language model decides when to call external tools, because that decision logic is central to building reliable agents. Instead of relying on prompt tricks, it uses JSON Schema function definitions in the same style as OpenAI function calls, and learns when to invoke one or several functions and how to read their results. That makes it a solid starting point for teams wiring an LLM into APIs or internal systems where structured output matters. The models were competitive on the Berkeley function-calling leaderboard, and the repo shows how to serve them through an OpenAI-compatible endpoint. Just keep in mind the project is deprecated and the code is an old snapshot, so treat it as a learning reference and architectural starting point rather than something to deploy fresh.