Hermes Agent: The Self-Improving AI Agent That Remembers Everything

Most AI agents have no memory. Every session starts from scratch — you re-explain your project, re-introduce your codebase, re-describe what you're working on. It's tedious, and it caps how useful the agent can actually become over time.

Hermes Agent, released by Nous Research under MIT license, is built around a different model: it learns your projects, creates skills from its own experience, and reaches you across Telegram, Discord, Slack, or WhatsApp from a single gateway. The longer it runs, the more capable it gets.

This guide covers installation, configuration, and the features that actually matter.

---

What Hermes Agent Actually Is

Hermes is a server-side autonomous agent — not a chatbot wrapper, not an IDE plugin. You install it on a Linux server, Mac, or WSL2 machine, connect your messaging accounts, and it runs persistently as a system service.

What sets it apart from tools like OpenClaw or standard LLM interfaces:

Persistent memory. Hermes indexes what it learns across sessions. It remembers how it solved a problem, stores project context, and retrieves it on demand. Ask it the same question two weeks later — it has the context from before.

Auto-generated skills. When Hermes solves a new type of problem, it encodes that solution as a reusable skill. Next time a similar task comes up, it pulls the skill instead of reasoning from scratch. Skills can be shared via agentskills.io or installed from communities on ClawHub and LobeHub.

40+ built-in tools. Web search, terminal access, file system, browser automation, vision, image generation, text-to-speech, code execution, cron scheduling, multi-model reasoning, and subagent delegation — all built in.

Multi-platform gateway. One installation, one gateway: Telegram, Discord, Slack, WhatsApp, Signal, Email, and CLI. Start a conversation on Telegram, continue it on Slack. Context carries over.

5 execution environments. Local, Docker, SSH, Singularity, and Modal — with container hardening and namespace isolation.

---

Requirements

• Linux, macOS, or WSL2 (Windows native is experimental and unsupported)
• No other prerequisites — the installer handles Python 3.11, uv, and all dependencies automatically

New to the terminal? The install command below is a single copy-paste. If you've never opened Terminal before, our Terminal Beginner's Guide covers exactly what you need.

---

Installation

One command installs everything:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

No sudo required. The script installs uv, Python 3.11, clones the repo, and sets up the virtual environment. Takes 2–5 minutes depending on your connection.

After install, run:

hermes setup

This launches an interactive wizard that walks you through connecting your model backend.

---

Choosing Your Model Backend

Hermes supports three backend options:

Option 1: Nous Portal (Recommended)

hermes setup
# Choose: Nous Portal
# Authenticate via OAuth at portal.nousresearch.com

Connects to Nous Research's hosted models, including their Hermes model series. Best experience, no API key management.

Option 2: OpenRouter

hermes model
# Choose OpenRouter
# Enter your OpenRouter API key

OpenRouter gives you access to 100+ models (GPT-4, Claude, Gemma, Mistral, etc.) through a single key. Useful if you want to switch models without reconfiguring Hermes.

Option 3: Any OpenAI-Compatible Endpoint

hermes model
# Choose: Custom endpoint
# Enter: http://localhost:11434/v1  (for local Ollama)

Running Ollama locally? Point Hermes at it. This means Hermes can orchestrate a fully local model — no cloud required.

Running heavier models locally? If you're running E4B-class models or doing batch inference, Ampere.sh offers affordable ARM-based GPU compute designed for AI workloads — significantly cheaper than AWS for inference tasks.

---

Setting Up the Multi-Platform Gateway

This is where Hermes goes from useful to powerful. One command launches the setup wizard:

hermes gateway setup

Walk through connecting whichever platforms you use. For Telegram (most common):

1. Create a bot via @BotFather → copy the API token
2. Paste into the Hermes gateway setup
3. Start the gateway:

hermes gateway

Or install it as a system service so it runs on startup and survives reboots:

hermes gateway install

After this, you can chat with Hermes on Telegram exactly like you'd chat with a person. It has full tool access — it can run terminal commands, search the web, generate images, transcribe audio, and execute code in response to your messages.

---

Natural Language Scheduling

One underrated feature: tell Hermes to do recurring tasks in plain English.

"Every Monday morning, summarize what I worked on last week and email it to me."
"At 9pm daily, check if any GitHub PRs are waiting for my review."
"Every Friday, run my SEO audit script and send me the results."

Hermes converts these to cron jobs and runs them unattended through the gateway. The equivalent of a personal automation layer that speaks your language.

---

Installing Community Skills

Hermes ships with 40+ bundled skills for MLOps, GitHub workflows, and research. Add more from the community:

hermes skill install <skill-url-or-github-path>

Browse available skills at agentskills.io or the ClawHub community.

Skills are stored in the open agentskills.io format — if Hermes figures out a useful workflow, it packages it and you can share it.

---

Keeping It Updated

hermes update

Pulls the latest code and reinstalls dependencies. Run anytime a new version drops.

---

Key Takeaways

• Install: Single curl command, no sudo, no prerequisites — works on Linux/Mac/WSL2
• Backends: Nous Portal, OpenRouter, or any local OpenAI-compatible endpoint (Ollama, LM Studio)
• Gateway: One deployment reaches Telegram, Discord, Slack, WhatsApp, Signal, and CLI
• Memory: Learns your projects and encodes solutions as reusable skills automatically
• Scheduling: Natural language cron — tell it what to do, when, in plain English
• License: MIT — fully open-source, free to use commercially
• GitHub: NousResearch/hermes-agent