MemGPT / Letta

Warnings

• breaking pip install memgpt is abandoned. Package renamed to letta in September 2024. All memgpt.* imports fail.
  Fix: pip install letta-client and update all imports to from letta_client import Letta
• breaking Letta agents are server-side resources, not in-process Python objects. You cannot instantiate an agent locally without a running Letta server.
  Fix: Start the server via Docker (letta/letta:latest) or use Letta Cloud before making any client calls.
• breaking Old memgpt API patterns (create_client(), Agent(), memgpt.run()) do not exist in the letta package. Architecture changed completely.
  Fix: Use letta-client SDK with client.agents.create() and client.agents.messages.create().
• gotcha letta (full package) and letta-client are different packages with different APIs. Most application code only needs letta-client.
  Fix: pip install letta-client unless you are running the server runtime itself.
• gotcha MemGPT as a term now refers only to the research paper's agent design pattern, not the framework. Searching 'MemGPT Python' returns a mix of abandoned pymemgpt, transition-era letta 0.x, and current letta-client code — all incompatible.
  Fix: Ignore any tutorial using memgpt package or memgpt.* imports. Use letta-client docs at docs.letta.com.

Install

• pip install letta-client Python (API client only, recommended)
• pip install letta Python (full server runtime)
• docker run -v ~/.letta/.persist/pgdata:/var/lib/postgresql/data -p 8283:8283 -e OPENAI_API_KEY=your_key letta/letta:latest Docker (recommended server deployment)

Imports

• Letta

  from letta_client import Letta

  memgpt package is abandoned. All imports from memgpt.* will fail.
• Letta (server)

  from letta import create_client

  Only use letta (not letta_client) if you are running the full server runtime locally.

Quickstart

Create a stateful Letta agent via the API client. Requires a running Letta server or Letta Cloud API key.

from letta_client import Letta
import os

client = Letta(api_key=os.getenv("LETTA_API_KEY"))

agent = client.agents.create(
    model="openai/gpt-4o",
    memory_blocks=[
        {"label": "human", "value": "Name: Alice"},
        {"label": "persona", "value": "I am a helpful assistant."}
    ]
)

response = client.agents.messages.create(
    agent.id,
    input="Hello, remember my name."
)
print(response.messages[0].content)