Can Some compare and explain its unique features when compared to flowise and similar projects agent ?

[Greatz08]

In Flowise and n8n and similar open source programs too we can create agents with memory so i wanna understand what all unique things this project provide so that we can utilize this project better. Those projects agents building is pretty easy and effective and they provide alottt of tools and other things so thats why i am curious about this one .

[mruderman]

I'd say it's the ability to have computer agents managing their own memory in a way that's completely transparent and initially influenceable by the human agent. It really depends on your use case, but this project will only become more unique and important as the idea of sentient computer beings becomes less and less of a fringe idea, especially, there's already I feel a slow creep from the concept of autonomous agents to a broader concept of autonomy in general, much the way that we think of it for ourselves. Memory is not singular, it operates at multiple levels, individual, collective, philosophical. In this way I see Letta as far less superficial. What seems important today, making money, completing tasks, the best 'automations' as you point out can be accomplished by other means and platforms. But in the long run Letta provides a path to more meaningful understanding and interaction with computer intelligence (while still being able to get stuff done and automate, myself included). Best of luck, I'm not formally affiliated with the project either by the way.

[cpacker]

Thanks for the kind comment @mruderman !

@Greatz08 sorry I missed this discussion thread earlier (I keep a closer eye on Discord and haven't been doing a great job at monitoring GitHub discussions). To basically paraphrase what @mruderman said - Letta is unique in that the high level focus of the project is around long-running agents, ie what is the system you need to build around LLMs so that you can build assistants/companions/workflows/automation that can run indefinitely and learn over time? Because of this, there's a very strong focus on memory in the project (one of the main limitations of agents today). Lots of other projects have primitive long-term memory (many of the projects directly using some of the original methods from Letta/MemGPT), but in Letta the entire agent runtime is optimized around the memory management concept. This focus also comes up in our API decisions - for example, in many agents APIs like the OpenAI Assistants API, there's a concept of "threads" (each conversation starts a "thread", which is generally intended to be ephemeral / not infinitely lasting) - instead in Letta there are no threads, just "agents" that maintain state across all usage indefinitely.

tl;dr we care the most about making long-running persistent agents (that learn over time like humans do), which influences how we design the APIs, design the UI / web app, write tutorials/cookbooks/example code, etc.