Letta
letta.com
Исследовательская лаборатория и платформа stateful-агентов, которые помнят всё и учатся со временем
Описание
Даёт stateful-агентов с постоянной памятью, которые можно запускать через CLI, десктоп-приложение или облако; в документации описаны MemFS, разделяемая память, подагенты, скиллы, моды, расписания, секреты и права доступа; агенты подключаются к каналам Slack, Telegram, Discord, WhatsApp и Signal, интегрируются с GitHub и ACP; поддерживаются облачные песочницы, Bring Your Own Machine и самостоятельный хостинг; есть Agent SDK и API-платформа с юзедж-биллингом. Letta Auto подбирает модель, работает BYOK с внешними ключами и кодинг-планами (ChatGPT/Codex, Kimi, Z.AI).
Letta is an AI platform for building stateful LLM applications with advanced memory capabilities. It allows developers to create, deploy, and manage AI agents at scale, offering a hosted cloud service, REST APIs, and tools for agent development. Letta emphasizes memory management, model-agnostic design, and transparent long-term memory for LLMs
Установка
npm i -g @letta-ai/letta-code| Ставится в | npm, cli, desktop-app, agent-sdk, rest-api, docker |
Не знаете, с чего начать — попросите ассистента провести по шагам:
Возможности
CLOUD DEPLOYMENT, MEMORY MANAGEMENT, MODEL-AGNOSTIC, WHITE BOX SYSTEMS, AGENT DEVELOPMENT ENVIRONMENT, REST APIS, PYTHON SDK, PERSISTENCE, TOOL CALLING
Сценарии использования
PERSONALIZED CHATBOTS, ENTERPRISE AI ASSISTANTS, CUSTOM TOOL INTEGRATION, AUTOMATED AI WORKFLOWS, LONG-TERM MEMORY AGENTS
Частые вопросы
Letta is an open-source framework for building stateful AI agents that can maintain memory and context across long-running conversations. Built by the researchers behind MemGPT, Letta enables developers to create AI assistants (chat or voice) that learn and evolve from interactions without starting from scratch each time.
Letta offers custom AI personalities, long-term memory, voice integration, REST APIs, privacy-first design, and open-source flexibility. The platform includes an advanced context management system that prevents agents from forgetting when their context window fills up, allowing them to maintain memories across sessions and continuously improve.
Letta uses a sophisticated memory hierarchy with self-editing memory split between in-context and out-of-context memory. The agent's in-context memory is composed of persistent editable memory blocks, and agents control their context window by using tools to edit, delete, or search for memory. This creates perpetual self-improving agents with infinite message history.
Letta integrates with multiple large language models including OpenAI, Anthropic, and Mistral.
Yes, Letta is designed for developers who have experience with language models. However, the platform provides developer quickstarts, guides, and educational resources like the Codecademy "Intro to AI Agents with Letta" course to help you get started.
You can deploy Letta using Railway's one-click deploy button. After deployment, you need to set environment variables (like OPENAI_API_KEY and DATABASE_URL), access the dashboard via Railway's provided domain, create your assistant, and integrate it with your apps.
Essential environment variables include OPENAI_API_KEY (to connect to an AI model), DATABASE_URL (to manage assistant state and user data), PORT (the network port where the app runs), and NODE_ENV (to define the environment as production or development).
You can build your first agent in 5 minutes using the Letta API with Python or TypeScript.
Letta supports a variety of applications including personalized chatbots requiring long-term memory and evolving personas (such as AI companions), agents connected to external data sources (like private enterprise ChatGPT deployments or medical assistants), agents connected to custom tools (such as chatbots that search the web for news), and automated AI workflows (like email monitoring agents).
Letta is completely free and open-source with no licensing fees or restrictions on usage. Your only costs come from hosting and API usage (such as OpenAI tokens or database hosting).
Letta provides full data ownership, open customization, and the freedom to choose your model—unlike closed platforms that lock you into subscriptions. The open-source nature means you can self-host the framework and maintain complete control over your agents and data.
Yes, Letta is open-source and self-hostable, giving you the option to run your own server rather than relying on Letta Cloud.
Характеристики
| Тип | Платформа |
| Категория | AI-агенты / Stateful-агенты и непрерывное обучение |
| Цена | есть бесплатный тариф (от $20/мес) |
| Платформа | Только macOS |
| Системы | cli, macos |
| Хостинг | hybrid |
| Установка | cli |
| Ставится в | npm, cli, desktop-app, agent-sdk, rest-api, docker |
| Для кого | Исследователи и разработчики AI-агентов, команды, строящие долгоживущих агентов с памятью, предприятия с требованиями SSO и ролевого доступа |
| Сложность | developer |
| Язык сайта | en |
| Вендор | Letta |
| Рейтинг | 0.00 (0 отзывов) |
| Просмотры | 94 180 |
| Запуск | 2024-11-14 |
