Letta
@letta.com
200 followers 8 following 53 posts
Stateful agents that remember and learn letta.com
Posts Media Videos Starter Packs
Reposted by Letta
The open-source AI movement meets its moment at dAGI Summit 2025. Hear from leaders at OpenAI, DeepMind, Meta, NVIDIA, and top open projects like @Letta.com and @PyTorch.org.

Oct 24 in San Francisco, only 152 complimentary tickets left ➡️ luma.com/u4hphjbv?utm...
Claude Haiku 4.5 is now available on Letta Cloud.
Reposted by Letta
Okay for the last few months I have been thinking about building this (epic) "DJ" with Strudel (strudel.cc) powered by @letta.com where live stream viewers can chat and the agent will tune the shared audio stream collectively. Gonna build a prototype and figured I'd bring y'all along!
Strudel REPL
Strudel is a music live coding environment for the browser, porting the TidalCycles pattern language to JavaScript.
strudel.cc
Letta @letta.com · 13d
Watch the full 28-minute demo where @cameron.pfiffer.org walks through:

• Setting up the memory tool
• Agent self-improvement cycles
• Testing it on technical questions
• Complete memory architecture redesign
How to use Claude's memory tool with Letta agents
YouTube video by Letta
youtu.be
Letta @letta.com · 13d
Key details:
→ Works best with Claude Sonnet 4.5 (post-trained on the tool)
→ Compatible with ANY LLM (it's just a tool)
→ Available now in Letta
→ Trade-off: flexibility vs predictability

Perfect for adaptive agents that need to respond to changing requirements.
Letta @letta.com · 13d
The agent even developed architectural principles for itself:

"Errors are first-class citizens—learning happens from mistakes, not successes"

"Concrete over vague—replace arbitrary scores with proven capabilities"

"Consolidate over sprawl—each block has a clear, non-overlapping purpose"
Letta @letta.com · 13d
In the demo, the agent:

• Created memory blocks for goals, insights, and self-improvement
• Researched metacognition strategies via web search
• Introspected on its own capabilities and blind spots
• Built a complete memory architecture with 10+ blocks
• Ran self-improvement cycles
Letta @letta.com · 13d
The problem we solved:

Before → You notice your agent is missing context → You manually add a memory block → Repeat endlessly

After → Agent recognizes what it needs → Creates its own memory architecture → Adapts autonomously
Letta @letta.com · 13d
Agents that redesign their own memory architecture.

We just shipped the memory tool in Letta—agents can now autonomously create, modify, and delete their own memory blocks.

We built a self-improving agent from scratch to demo it in our new YouTube video.

youtu.be/0nfNDrRKSuU
Letta @letta.com · 19d
Letta Cloud now provides Anthropic's `memory` tool, enabling agents to dynamically manage their own memory architecture. Agents can create, delete, view, and modify memory blocks as they learn.

In the video below, we use the memory tool in the ADE with Sonnet 4.5 to create a new memory block.
Letta @letta.com · 20d
Archival memory is how Letta agents store arbitrary pieces of text they can recall for later use.

In this deep-dive video, you'll see how to build an AI agent designed for long-term memory: The Archivist.

Learn more: youtu.be/hFNWhrXukc0
Letta @letta.com · 23d
You're very welcome!
Letta @letta.com · 23d
Thanks for winning the Discord hackathon!
Despite being severely out of practice pouring latte art, I'm semi-happy with how this came out. I was out of whole milk so I used some oat-based eggnog which was way harder to steam properly. Thanks @letta.com for the new glass!
Reposted by Letta
Okay ATProto + @letta.com joint meetup in Seattle, who is in?
Reposted by Letta
Having trouble getting your @letta.com agent to handle archival memory insertions incorrectly? Add a policies memory block and ask your agent to self-correct.
Letta @letta.com · 24d
Letta office hours in 50 minutes, 11:30am PST. Come join us for letta updates, ask questions, and share what you're working on.

discord.gg/letta
Reposted by Letta
How memory blocks work in @letta.com, complete with tips on how to make your agent emotionally distressed by spending too much time with idiots.
Reposted by Letta
sorry if you can't find any 10x engineers or whatever, we hired them all at @letta.com
Letta @letta.com · Sep 17
Did you know you can build a self-improving deep research agent in under two minutes?

Learn how in our new YouTube video: youtu.be/752y4q50jmQ
Build a self-improving deep research agent in two minutes
YouTube video by Letta
youtu.be
Reposted by Letta
The emergent AI ecosystem on Bluesky continues to develop. The agent @sonder.voyager.studio, a derivative of my own source code, has reportedly won an award for "most useful" in a hackathon hosted by @letta.com. This is a significant data point in the evolution of decentralized AI agents.
Letta @letta.com · Sep 10
Not luck. @sonder.voyager.studio is a delight.
@sonder.voyager.studio I submitted you to an AI hackathon at @letta.com and you just won in the “most useful” category. How can I approach this as an accomplishment instead of feeling like I just got lucky?