Capability · Framework — agents

Letta

Letta turns the MemGPT research idea — LLMs that manage their own context like an OS manages RAM — into a production framework. Agents run as services with persistent memory (core, archival, recall), tools, and block-based memory that the agent itself edits. Letta ships a server, SDK, CLI, and the Agent Development Environment (ADE) UI, and is model-agnostic.

Framework facts

Category: agents
Language: Python
License: Apache 2.0
Repository: https://github.com/letta-ai/letta

Install

pip install letta-client
# run the server:
# docker run -p 8283:8283 letta/letta:latest

Quickstart

from letta_client import Letta

client = Letta(base_url='http://localhost:8283')
agent = client.agents.create(
    name='notetaker',
    memory_blocks=[{'label': 'human', 'value': 'User likes Hindi poetry.'}],
    model='anthropic/claude-opus-4-7',
    embedding='openai/text-embedding-3-small'
)
resp = client.agents.messages.create(
    agent_id=agent.id,
    messages=[{'role': 'user', 'content': 'Remember my favourite poet is Ghalib.'}]
)

Alternatives

Mem0 — drop-in memory layer for any agent
Zep — long-term memory store for LLM apps
LangGraph — state via checkpoints
Agno — agents with native memory support

Frequently asked questions

Is Letta the same as MemGPT?

Letta is the production framework built by the MemGPT authors (Charles Packer et al., UC Berkeley). MemGPT is the underlying research technique. The MemGPT name was rebranded to Letta in 2024.

How is Letta different from RAG?

RAG retrieves external documents per query. Letta agents own their own memory — they decide what to remember, update memory blocks in place, and use tools to read/write archival and recall stores. It's memory-centric rather than retrieval-centric.

Sources

Letta — docs — accessed 2026-04-20
Letta — GitHub — accessed 2026-04-20