letta

letta

Letta is a leading agent framework in the 2026 AI agent engineering landscape (23.6k★).

When to use this skill

Stateful agents with self-editing memory (MemGPT research lineage)
Memory blocks (e.g. human/persona) the agent rewrites via memory tools
Letta Code CLI: local terminal agent with skills + subagents + continual learning (npm i -g @letta-ai/letta-code)
Letta API/server: REST + Python/TS SDKs for app integration
Model-agnostic; model leaderboard published (leaderboard.letta.com)
Agent File (.af) portable agent format; sleep-time compute for background memory maintenance
Ecosystem overlap: agent-memory-architecture (this repo) covers memory design concepts and already references Letta as a backend — letta skill covers operating it concretely

Installation

npm install -g @letta-ai/letta-code   # CLI; server SDKs: pip install letta-client / npm i @letta-ai/letta-client

Core API

import Letta from "@letta-ai/letta-client";
const client = new Letta({ apiKey: process.env.LETTA_API_KEY });
const agentState = await client.agents.create({
  model: "openai/gpt-5.2",
  memory_blocks: [{ label: "human", value: "Name: Timber..." }],
  tools: ["web_search", "fetch_webpage"],
});

Instructions

Step 1: Initialize the environment

Verify you have Python installed and set up properly. Use the correct package manager (pip, npm, pnpm, uv) as specified in the installation section.

Step 2: Establish the agent flow

Architecture pattern: Agent = model + memory blocks + tools persisted server-side; memory edited by the agent itself (self-improving); ADE (Agent Development Environment) for inspection

Step 3: Implement tool integrations and policies

Integrate necessary tools, environment keys, and safety gates according to your architectural needs. Position in the ecosystem: The stateful-memory agent lane; alternatives: mem0/graphiti (memory layers without agent runtime), deepagents (harness without self-editing memory)

Examples

Example 1: Minimal Agent Run

import Letta from "@letta-ai/letta-client";
const client = new Letta({ apiKey: process.env.LETTA_API_KEY });
const agentState = await client.agents.create({
  model: "openai/gpt-5.2",
  memory_blocks: [{ label: "human", value: "Name: Timber..." }],
  tools: ["web_search", "fetch_webpage"],
});

Best practices

Follow the primary installation commands for Letta to ensure dependencies are resolved correctly.
Configure model providers and API keys securely before starting the runtime.
Use appropriate sandboxing or permission isolation if executing AI-generated code.

About this skill