[Discussion] SPADE and LLM-based Agents (using FIPA)

[siebert-julien]

Hi everyone,

first of all I am new to Spade so forgive me if I am asking newbie questions and as the title says it is more a discussion than an issue (I could not find any discussion group).

I am interested in using FIPA ACL and (potentially) BDI architecture in relation to LLM.
I did my PhD a long time ago on multiagent oriented programming and multiagent-based simulations and nowadays I am leading a couple of projects using LLM. I am currently looking at multiagent platform in the LLM community (Autogen, MetaGPT, CrewAI, etc.) and I am missing all the research done on MAS > 1990s (like ACL, BDI, etc.).

I was wondering:

1. are there any examples of MAS build with Spade that uses LLM as (part) of the agent behaviour?

2. [not related to LLM], when sending messages to other agents in Spade, are the messages following the FIPA ACL specification? For instance I'd like to use Communicative Acts (like Accept, Cancel, Request, etc. see http://www.fipa.org/specs/fipa00037/SC00037J.html#_Toc26729690), does Spade support this (I could not find anything in the documentation) or do I have to build this myself?

Best, Julien

[javipalanca]

Hi Julien,

Currently, introducing LLMs into the agent world and agents into the LLM world is very trendy. You should expect results on SPADE soon ;)

Regarding FIPA, although the early versions of SPADE did comply with the full FIPA standard, currently, and mostly due to the low FIPA activity, SPADE has focused more on complying with the XMPP standard.
To use communicative acts, you can insert metadata into messages. E.g.: {“performative”: “inform”}