Over the last year we've had a lot of innovation in the multi-agent AI space, with folks like
@CamelAIOrg
leading the charge, my work with agent-actors (below),
@yoheinakajima
's BabyAGI,
@pyautogen
, and recently
@joaomdmoura
's CrewAI.
So what comes next?
Most projects using agents have used one agent. Or, if theres been many agents, they typically interact in a sequential manner
Excited to share a new guest blog post from
@cyrusofeden
that pushes the boundary - multiple agents acting in parallel
I'm thrilled to announce the alpha release of my labor of love — Llegos — a domain specific library for exploring the design space multi-agent systems!
One of our early users called it "more of a PyTorch, less of a Keras".
What's the deal with Llegos?
Existing multi-agent projects all provide a *specific implementations of agent interaction patterns*.
Llegos changes the game — it makes it easier to experiment with different implementations of agent interaction patterns.
Said another way, you could use Llegos to express my designs in CamelAI, agent-actors, AutoGen, CrewAI — but not vice versa.
The best part? Llegos is compatible with all the above.
How??
Llegos strives to do one thing well — being the communication fabric in multi-agent systems.
Llegos lets you go from a diagram like this, to code like that.
Looking for collaborators, DM's open!
I'm excited to see what incredible systems the community will create with Llegos. Let's build the future together!
And for those wondering what's on the roadmap...
- Refining Developer Experience based on feedback from developing systems (until we get reasonably stable)
- A llegos.development module with asyncio actors for boosting performance
- A production module using
@raydistributed
…