I have implemented an architecture I call the Unit. It is a combination of 2 assistant and 2 threads, one is conversational in nature and only talks with the user, the other is functional and perform all real world manipulation and interacts with JSON schema only. This is what provided my results and also I believe is close to the minimum viable agent. And conversational, functional, auto updating unit with memory.
I will make a post on the architecture of the unit soon. But his is here so people can finally replicate my results and hopefully fix some instabilities
I have decided to upload my PR for the post that gained traction https://github.com/daveshap/OpenAI_Agent_Swarm/discussions/62.
I have implemented an architecture I call the Unit. It is a combination of 2 assistant and 2 threads, one is conversational in nature and only talks with the user, the other is functional and perform all real world manipulation and interacts with JSON schema only. This is what provided my results and also I believe is close to the minimum viable agent. And conversational, functional, auto updating unit with memory.
I will make a post on the architecture of the unit soon. But his is here so people can finally replicate my results and hopefully fix some instabilities