How to Self-Host Letta with Docker: Step-by-Step Setup

Want your own persistent AI agent with memory — without sending data to a third-party cloud? Self-hosting Letta with Docker takes about ten minutes, and the official letta/letta:latest image makes it painless. Here's the complete walkthrough.

Step 1: Pull and run the container

Create a directory for persistent data, then start Letta with a single command:

mkdir -p ~/letta-data && cd ~/letta-data
docker run -d --name letta \
  -p 8080:8080 \
  -v $(pwd)/data:/data \
  letta/letta:latest

The web UI is now available at http://localhost:8080. The /data volume keeps your agents' memory safe across container restarts and upgrades.

Step 2: Connect your LLM

In the settings panel, add an API key for OpenAI, Anthropic, or any OpenAI-compatible endpoint. Running a local model? Point Letta at Ollama and you get a fully offline agent stack — memory included.

Step 3: Create your first agent

Use the ADE's agent builder to give your agent a name and a persona. As you chat, watch the memory inspector update in real time — the agent is writing its own memory blocks. Ask it to remember facts, and verify them in a new session. That persistence is the whole point.

Hardware expectations

That's it — a stateful agent with self-editing memory, fully under your control. For compose templates, port notes, and upgrade tips, check the tool page below.