[HN Gopher] Letta: Letta is a framework for creating LLM service...
       ___________________________________________________________________
        
       Letta: Letta is a framework for creating LLM services with memory
        
       Author : sebg
       Score  : 103 points
       Date   : 2025-03-07 21:33 UTC (1 days ago)
        
 (HTM) web link (github.com)
 (TXT) w3m dump (github.com)
        
       | jxjnskkzxxhx wrote:
       | What's an "LLM with memory"? Is it an agent that knows how to
       | insert into, and query, a database?
        
         | blooalien wrote:
         | https://docs.letta.com/letta-platform
        
         | pacjam wrote:
         | In Letta that's referring to an LLM agent that has persistent
         | memory (eg keep running the agent far beyond its context window
         | limit, but maintain some persistent state to avoid derailment /
         | enable indefinite use / learning over long horizons).
         | 
         | If you implement this persistent memory via tool use, then it's
         | related to inserting/querying databases, because some agent has
         | to read/write to a data store (potentially backed by a
         | database) to maintain memories.
         | 
         | -charles from letta
        
       | rcarmo wrote:
       | I find it somewhat amusing that Lethe was the river of
       | forgetfulness, and that the new name for the project is so close
       | to that.
        
         | teleforce wrote:
         | Thanks for the info, never heard of Lethe before [1].
         | 
         | Fun facts in Arabic, "insan" meaning human being or mankind.
         | This word takes its root from the word "nisyan", which
         | literally means "forgetfulness".
         | 
         | [1] Lethe:
         | 
         | https://en.wikipedia.org/wiki/Lethe
         | 
         | [2] Arabic for forgetfulness:
         | 
         | https://arabic.fi/words/3766
        
           | andsoitis wrote:
           | > This word takes its root from the word "nisyan", which
           | literally means "forgetfulness".
           | 
           | There are two linguistic theories: the one you mention, and
           | the other which is from "uns" ('ns) which means
           | "companionship" or "affection."
           | 
           | While both explanations have historical significance, the
           | stronger linguistic evidence points to the latter explanation
           | (companionship or sociability) by Classical Arabic linguists.
           | The former (forget) is a more recent interpretation linked to
           | philosophical and theological discussions rather than
           | libguistic roots.
        
       | chrisvalleybay wrote:
       | I didn't find any thing on Letta and MCP. Any thoughts on
       | integrating MCP into the platform?
        
         | anon373839 wrote:
         | Totally serious question: is MCP basically an in-joke right
         | now? I mean, I know what it is. I just don't know why I'm
         | seeing it everywhere in the last 3 days?
        
           | imtringued wrote:
           | Maybe this will help you figure it out:
           | 
           | https://github.com/modelcontextprotocol/servers
        
             | yard2010 wrote:
             | Thank you for the mind-blowing 2 hr rabbit hole. I'm going
             | to use this in my personal helper bot.
        
           | chrisvalleybay wrote:
           | Probably people are excited about the prospect of having this
           | problem be solved so that we can focus more on getting the
           | agents to work rather than all be reimplementing the same
           | data fetchers/parsers etc.
        
             | anon373839 wrote:
             | Sure, but I mean why so suddenly? This has been out a
             | while...
        
               | _neil wrote:
               | I think it's just reached critical mass (and has been
               | picked up by the hype peddlers). It's been gaining
               | momentum for months.
        
           | Fabricio20 wrote:
           | Interesting question, I personally first saw it because
           | jetbrains now has a plugin for it on all their IDEs.
        
           | pacjam wrote:
           | I'm also not really sure what's going on with the hype wave
           | atm (people like having things to hype?) since MCP has also
           | been around for a bit, seems like a bit of a meme.
           | 
           | IMO, the main value (as an agent / AI developer) that you get
           | from MCP is another style of tool repository. Similar to how
           | it's nice to have pre-made tools (eg Composio), MCP servers
           | are another repo of tools to pull from (you just need to add
           | MCP client functionality to your agent loop). So if a ton of
           | energy is going into building MCP servers, might as well add
           | MCP client support so that you can take advantage of the
           | additional tools.
           | 
           | -charles from letta
        
           | irickt wrote:
           | Tangential cultural reference ...                 https://en.
           | wikipedia.org/wiki/List_of_Tron_characters#Master_Control_Pro
           | gram            The Master Control Program (MCP) ... is the
           | main digital antagonist of the first Tron film.
        
         | pacjam wrote:
         | We're working on it, I'm hoping we can get an integration in
         | sometime next week! If you want to track progress on it would
         | recommend joining our Discord.
         | 
         | To clarify - by "it", I mean making Letta agents an MCP client,
         | so that you can easily connect to MCP servers as tools (not the
         | other way around, where Letta itself is an MCP server - if you
         | want that, someone in the community already made one:
         | https://github.com/oculairmedia/Letta-MCP-server)
         | 
         | -charles from letta
        
       | redwood wrote:
       | This was the berkeley team that previously released MemGPT
        
         | pacjam wrote:
         | indeed - hello there
        
       | dangbeetle wrote:
       | My father is suffering from increasingly bad short term memory.
       | This has been a slow process that has taken many decades. My mom
       | was worried about his memory 20 years ago. Now he lives alone and
       | has nobody to help remind him of things.
       | 
       | He is highly functional, able to multiply two digit numbers in
       | his head and run heavy equipment on his farm. But he'll ask you
       | the same question, five minutes after he asked it the last time.
       | He'll forget what neighbor just visited him the day before, or
       | whether he called the gas company to refill his propane barrel.
       | 
       | I believe that a voice interactive LLM with memory could be
       | extremely helpful to him and to other elderly people. He could
       | use it as a daily log, telling it things that have occurred, and
       | then interrogate it later. In truth, who among us could not
       | benefit from such a system, even in our younger years?
       | 
       | This framework looks interesting to use for such a project. Are
       | there any other people working on similar things?
        
         | pacjam wrote:
         | > "voice interactive LLM with memory"
         | 
         | Our team has been working really hard on a low-latency voice
         | integration that enables (open source) "advanced voice mode",
         | but with long-term memory!
         | 
         | If you're interested definitely hop in our Discord to keep
         | track of dev progress (you can also monitor our docs). It's all
         | OSS of course, so you can use it with any model you want and
         | run it all locally (your agent data stays on device).
         | 
         | Once available you can definitely use Letta for this usecase -
         | basically provision a stateful Letta agent for your dad, with
         | instructions / prompting that your dad has memory loss issues,
         | and the agent should be proactive in recording memories (and
         | potentially jogging your dad's memory too). All the
         | memory/state will be stored in postgres, so you can expose it
         | to other tools as well, eg could even collate the memories into
         | a nextjs/mobile app where you dad can browse them.
         | 
         | -charles from letta
        
       ___________________________________________________________________
       (page generated 2025-03-08 23:01 UTC)