Self-learning AI agent Hermes: capabilities and server deployment
Nous Research
OpenAI
Anthropic
Google/DeepMind
DeepSeek
Hermes Agent from Nous Research is a self-learning AI assistant that retains context across sessions, stores facts and skills, and can be deployed on a remote server. The article explains its memory system, skill creation, and practical launch in Amvera Cloud.
Hermes Agent by Nous Research is an AI agent that goes beyond single-turn chatbots by preserving memory, session history, and reusable skills. It uses large language models like GPT, Claude, Gemini, and DeepSeek as its brain, and can execute terminal commands, search the web, use a browser, and connect to MCP servers. The agent stores short facts in MEMORY.md and USER.md files, and performs full-text search across past sessions. Skills are saved step-by-step instructions for repetitive tasks, stored as SKILL.md files. Hermes can also schedule cron tasks, delegate work to sub-agents, and interface with Telegram, Discord, Slack, and more. The article provides a step-by-step guide to deploying Hermes on Amvera Cloud using a Docker image, requiring a Telegram bot token, an OpenRouter API key, and a SERPAPI key for web search. The minimum hardware is 1 GB RAM and 1 CPU core, with 2-4 GB recommended. Safety precautions include restricting access via TELEGRAM_ALLOWED_USERS and avoiding HERMES_YOLO_MODE.
- Сокращения
- LLM = Large Language Model — большая языковая модель
- CLI = Command-Line Interface — интерфейс командной строки
- MCP = Model Context Protocol — протокол контекста модели
- VPS = Virtual Private Server — виртуальный частный сервер
- VDS = Virtual Dedicated Server — виртуальный выделенный сервер
- ID = Identifier — идентификатор
- API = Application Programming Interface — интерфейс программирования приложений
Source: Habr — хаб ML —
original