Stateful agents with durable memory

Letta Starter

A compact Python starter for Letta agents that remember across sessions, manage their own memory blocks, and run inference through Nebius Token Factory.

Persistent Memory survives CLI restarts
Minimal One Python file plus Docker setup
Portable OpenAI-compatible Nebius endpoint

Agent memory

Built around Letta core memory blocks.

Human block

Stores useful facts about the user, such as identity, preferences, goals, and projects.

Persona block

Defines the agent personality so the assistant remains consistent across conversations.

Self editing

Lets the agent update memory as it learns, then reload the same state on the next run.

Quick setup

Run the server, install the client, start the CLI.

1. Letta server
cp .env.example .env
# edit NEBIUS_API_KEY
docker compose up -d
2. Python client
pip install -r requirements.txt
# or: uv sync
3. Interactive CLI
python main.py

You Hi, my name is Arindam and I love building AI apps.

Sam Nice to meet you, Arindam. I will remember that.

Restart python main.py

You What's my name and what do I like to build?

human memory

The human's name is Arindam. They love building AI apps.

Open source

Clone it, run it, and make it yours.

The repository includes the Python CLI, Docker configuration, environment template, quickstart docs, and this deployment-ready website.

Open on GitHub