---
id: letta
version: "0.16.8"
license: Apache License
license_treatment: permissive
maintenance: active
---

# letta — Create LLM agents with long-term memory and custom tools

License: permissive · Maintenance: active · Downloads: 148.7K/mo

## What it is and what it does

Letta is a framework for creating AI agents that maintain long-term memory and can be customized with tools and skills. It offers two main entry points: a local CLI tool (Letta Code, installed via npm) that runs agents on your machine, and a Python API (letta-client) for embedding stateful agents into applications. The framework handles agent state management, memory persistence, and tool integration, allowing agents to learn and adapt over time.

The package itself (letta) is the backend service and core framework, while letta-client is the Python SDK for API interaction. It supports multiple LLM models and is model-agnostic, though the documentation recommends specific models for optimal performance. The dependency footprint is substantial—69 runtime packages including integrations with Anthropic, OpenAI embeddings, LlamaIndex, and observability tools like Datadog—reflecting its role as a full-featured agent platform rather than a lightweight library.

Use it for:

- Build conversational AI assistants that remember user context and preferences across sessions.
- Create autonomous agents that can use web search, fetch webpages, and other tools to complete tasks.
- Integrate stateful LLM agents into existing applications via the Python SDK with persistent memory.
- Develop self-improving agents that learn from interactions and adapt their behavior over time.
- Run local AI agents in the terminal for coding assistance and task automation using Letta Code.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Letta is a framework for building LLM agents with persistent memory that can learn and improve over time, accessible both as a local CLI tool and through a Python API for integration into applications.

Yes, if you need a production-grade agent framework with memory management and tool integration. The active maintenance, permissive license, and low install friction support adoption. However, the large dependency tree and requirement for an external API key (for the cloud API) or Node.js 18+ (for the CLI) may be friction points. Start with letta-client if you only need the Python SDK; the full letta package is for running the backend service.

## Install

pip install letta
uv add letta
poetry add letta

## Installing letta

Before you install: Low install friction with a pure-Python wheel distribution. Actively maintained with a recent release. Requires Python 3.11 or later and brings in a large dependency tree (69 runtime packages including anthropic, llama-index, and datadog), which may add setup complexity in constrained environments.

License in practice: Licensed under Apache License with permissive treatment, allowing commercial and private use with minimal restrictions.

Quickstart:

pip install letta

from letta_client import Letta
import os

client = Letta(api_key=os.getenv("LETTA_API_KEY"))
agent = client.agents.create(
    model="openai/gpt-5.2",
    memory_blocks=[{"label": "human", "value": "Name: Alice"}],
    tools=["web_search"]
)
response = client.agents.messages.create(
    agent_id=agent.id,
    input="What do you know about me?"
)

Requires a Letta API key (from app.letta.com) and Python 3.11 or later. The CLI tool also requires Node.js 18+ if using the Letta Code command-line interface.

Verify before relying:

- Whether all 69 runtime dependencies are required for basic usage or if many are optional for specific integrations.
- Performance characteristics and latency when managing agent state and memory across requests.
- Compatibility details with specific LLM providers beyond the examples shown (Anthropic, OpenAI).

## Package facts

- License: Apache License (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 148.7K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

llm agents with memory, stateful ai agents, long-term memory for ai, self-improving agents, agent framework python, ai memory management, agent api integration, agent-framework, memory-management, llm-integration

[View on SkillFed](https://skillfed.io/packages/letta) · [View on PyPI](https://pypi.org/project/letta/)