Letta – Stateful AI Agents, MemGPT & Persistent Memory Platform | Quasa.io

#Quasa #QUA #Letta

Letta is the company behind MemGPT. It builds agents that keep state between sessions instead of starting from a blank prompt. Memory is split into core blocks the agent can edit, archival memory for long facts, and recall for past turns. The agent moves data in and out of the context window with tools, not a hidden vector dump. Letta Code is the coding agent. The platform adds an agent development environment, an API, scheduled tasks, and channels such as Slack, Discord, and Telegram. You can bring your own model keys. A free tier is commonly cited with a small agent cap; paid plans have been listed around $20 a month plus usage. Confirm the live SKU. Named users include Bilt and 11x.

𝐂𝗢𝗥𝗘 𝗦𝗧𝗥𝗘𝗡𝗚𝗧𝗛𝗦
• Memory you can open and edit, not a black box
• Same agent across sessions and model swaps
• Self-editing core memory plus archival search
• Open research lineage from the MemGPT paper
• API and self-host, not only a chat tab  

𝗜𝗗𝗘𝗔𝗟 𝗙𝗢𝗥
Teams that need an agent to remember a user, a codebase, or a process, and to show what it stored.

𝗛𝗜𝗚𝗛𝗟𝗜𝗚𝗛𝗧𝗦
• Core, archival, and recall as separate stores
• Sleep-time consolidation of old context
• Git-tracked agent files such as skills and prompts
• Multi-channel delivery
• BYOK so model cost stays on your account  

𝗣𝗢𝗧𝗘𝗡𝗧𝗜𝗔𝗟 𝗖𝗢𝗡𝗦𝗜𝗗𝗘𝗥𝗔𝗧𝗜𝗢𝗡𝗦
• The agent decides what to keep; it can drop a fact
• Free caps are real; active agents add a meter
• Inspectable memory is not the same as correct memory
• Younger product surface than the paper’s fame
• A remembered preference is not a permission  

𝗢𝗩𝗘𝗥𝗔𝗟𝗟 𝗩𝗘𝗥𝗗𝗜𝗖𝗧
4.3/5 stars. Use Letta when the job is “this agent should still know us next week.” Do not treat a memory block as a system of record. Earn 1 QUA reward by reviewing on Quasa.io too!

𝗚𝗘𝗧 𝗦𝗧𝗔𝗥𝗧𝗘𝗗: https://quasa.io/projects/letta