Hermes AI Assistant
Hermes — Your AI Assistant That Actually Remembers
Hermes Agent by Nous Research is an open-source AI assistant with multi-level persistent memory and 40+ built-in tools. Run it on Telegram, Discord, WhatsApp, WeChat, or web chat. Managed hosting from $3/month — no VPS, no Docker.
Why Hermes vs Generic AI Chat
Most AI assistants — ChatGPT, Claude, Gemini — forget you the moment a chat closes. Each new session starts cold. You re-explain context every time. For a real assistant that builds on what it learned yesterday, you need persistent memory.
Hermes Agent has multi-level memory built into the core: short-term session memory, file-level memory the agent writes to as it works, and a long-term knowledge store it can recall from months later. The agent gets more useful the longer you use it — the opposite of how most chatbots feel.
And it lives where you live. Telegram, Discord, WhatsApp, WeChat, web chat — same agent, same memory, same tools across every channel.
What You Get
Multi-Level Persistent Memory
Hermes remembers across sessions, channels, and months. File-level memory, session memory, and a long-term knowledge store the agent reaches into when relevant. Most "AI assistants" forget you the moment a conversation ends.
Multi-Channel Out of the Box
Telegram, Discord, WhatsApp, WeChat, and web chat all work without webhook setup. Send a message on any of them and Hermes replies with the same memory and tools.
Bring Any Model
Anthropic Claude, OpenAI GPT, OpenRouter, Google Gemini, DeepSeek, Qwen, Hermes 3/4, or local models via Ollama and llama.cpp. Hermes is model-agnostic by design.
40+ Built-in Tools
Web search, terminal access, browser harness, vision, code execution, scheduled tasks, MCP server support, and subagent delegation. Plus skills and plugins from the upstream community.
Open Source, MIT License
No vendor lock-in. The full Hermes Agent source is on GitHub. Self-host on your own VPS, or use managed hosting and migrate later if you change your mind.
Managed Hosting From $3/mo
Skip the VPS, Docker, and SSL setup. Hermes models run on managed infrastructure today; native Hermes framework hosting is in private beta. From $3 first month, then $6/mo flat.
Common Use Cases
Personal AI Assistant
A private chat companion that learns your preferences, remembers context across weeks, and reaches you wherever you message — Telegram on phone, Discord on desktop, web chat at home.
Team Knowledge Bot
Drop Hermes into your team Discord. Feed it your docs and runbooks. New hires can ask "how do I deploy" and get a real answer with links — without pinging an engineer.
Research Companion
Web search + persistent memory makes Hermes a research partner that builds context over a multi-week project, not a chat that forgets your topic between sessions.
Customer Support Bot
Run Hermes on Telegram or WhatsApp with your product knowledge. It handles tier-1 support, remembers customers across messages, and escalates only when needed.
Self-Host or Managed?
Both are first-class paths.
- Self-host on your own VPS for full data sovereignty and the option to run local models. Expect ~30 minutes of setup and ongoing maintenance. See install guide, Hermes on Hetzner, or Hermes on Hostinger VPS.
- Managed on OpenClaw Launch from $3/mo. Hermes models run on managed infrastructure today; native Hermes framework hosting is in private beta. Auto SSL, backups, monitoring, 99.9% uptime.
How Hermes Compares
Several AI assistants pitch “persistent memory” in 2026. Most ship a shallow version — one extra context window, no real cross-session recall, no multi-channel sync. Hermes is one of the few that designed the memory architecture into the core agent loop, not bolted on. See the comparisons:
Privacy + Data Sovereignty
Hermes is open source under MIT. Your conversations live on infrastructure you choose — your VPS, or our managed servers with AES-256-GCM encryption at rest. We never read your messages and we don't train on user data.
For full air-gapped operation, run Hermes locally with Ollama or llama.cpp. The agent and the model both run on your hardware; no API calls leave your machine.
Get Started
Managed (10 seconds): Sign up at /register, pick Hermes from the configurator, paste your bot token, click deploy. Live in 10 seconds. From $3/mo first month.
Self-hosted (30 minutes): Provision a VPS (Hetzner CPX21 recommended), install Docker, pull ghcr.io/nousresearch/hermes-agent:latest, paste a config, run. Detailed steps in our install guide.
Pricing
Hermes Agent itself is free (MIT license). You only pay for hosting and inference:
- Self-host: ~€5–15/mo VPS + your time + model API costs
- Managed Lite: $3/mo first month, $6/mo after — AI credits included
- Managed Pro: $20/mo — multi-instance, higher quotas, more credits
- Local model inference: $0 model cost, but you pay in hardware (RAM, ideally GPU)
Full pricing details.
Bottom Line
Hermes is the AI assistant for people who want memory that survives the next session, deployment that survives a model swap, and infrastructure they can move. Run it on managed hosting today — or self-host whenever you're ready.