One-Sentence Answer
Hermes Agent is an open-source AI assistant that runs on your own computer, remembers who you are across sessions, connects to your messaging apps, and works with any AI model — making it more like a persistent digital team member than a disposable chat window.
The Problem: AI Tools That Forget and Can't Act
Most people use AI through one of three interfaces today:
- ChatGPT — a web or mobile chat window that forgets almost everything between conversations
- Claude Code — a terminal-only coding agent tied to Anthropic's models
- OpenAI Codex — a cloud-hosted coding agent tied to OpenAI's models
Each of these tools is good at what it does. But they share three limitations that make them feel like rented assistants rather than team members:
- No long-term memory. Start a new conversation and you are explaining your project, preferences and setup all over again.
- Locked to one provider. Switch from GPT to Claude to a local model and you need a different tool entirely.
- Trapped in one interface. ChatGPT lives in a browser tab. Claude Code lives in a terminal. Neither is reachable via Telegram at 2am when you have an idea.
Hermes Agent, built by Nous Research, was designed to solve all three problems at once.
What Is Hermes Agent?
Hermes Agent is an open-source AI agent framework that runs on your own machine — Windows, macOS, or Linux. It is not a website you visit or an app you sign into. It is a process that lives on your computer and makes itself available wherever you are.
Concretely, Hermes Agent can:
- Run in your terminal as an interactive chat
- Launch a native desktop app with a full GUI
- Connect to Telegram, Discord, Slack, WhatsApp, Signal, iMessage, Teams and 15+ other platforms — with the same tools available everywhere
- Work with any LLM provider: OpenAI, Anthropic, Google, Nous Portal, DeepSeek, xAI, local models, and 20+ more
- Execute commands on your system — run git, manage files, call APIs, browse the web, generate images
- Remember your preferences, project details, environment setup and past decisions permanently
- Run scheduled tasks (cron jobs) that deliver results to your phone every morning
- Get smarter over time by saving skills — reusable procedure documents it loads when relevant
The key mental shift is this: Hermes is not a chatbot you use. It is a system you set up once and then interact with daily, from whichever surface is most convenient at that moment.
Hermes Agent vs ChatGPT vs Claude Code vs Codex
Here is a direct comparison across the dimensions that matter most for daily use:
| Feature | ChatGPT | Claude Code | OpenAI Codex | Hermes Agent | |---------|---------|-------------|--------------|--------------| | Runs on | OpenAI cloud | Your terminal (Anthropic API) | OpenAI cloud | Your machine (local process) | | Models | GPT-5, o3/o4 | Claude 4 only | GPT-5 only | Any provider, switch mid-session | | Memory | Per-thread, limited | None between sessions | Per-task | Persistent, cross-session | | Messaging | Web/app only | Terminal only | Web dashboard | Telegram, Discord, Slack, 20+ | | Scheduled tasks | No | No | No | Yes (cron, webhooks) | | Code editing | Via artifacts/files | Excellent (native) | Excellent (native) | Excellent (via Claude Code/Codex sub-agents or directly) | | Open source | No | No | No | Yes (MIT) | | Local model support | No | No | No | Yes (Ollama, llama.cpp) | | Cost | ~$20/mo subscription | ~$20/mo + API usage | ~$200/mo (Pro) | Free agent + pay for API tokens |
The short version: ChatGPT is a general assistant. Claude Code and Codex are coding specialists. Hermes is an open platform that can do both — and extends into messaging, automation and personal workflows those tools were never designed for.
What Makes Hermes Genuinely Different
The comparison table above captures the feature differences, but three design choices are worth calling out because they change how you work with AI day to day.
1. Skills: An Agent That Gets Smarter Over Time
When Hermes solves a complex problem, discovers a workflow, or gets corrected by you, it can save that knowledge as a skill — a structured document describing the procedure, tools needed, known pitfalls and best practices.
Skills load automatically in future sessions when the agent recognises a relevant task. This means:
- The first time you deploy a Next.js site through Hermes, you walk it through each step.
- The second time, it already knows your deploy pipeline, environment variables and verification checklist.
- By the tenth time, the deploy is one message: "deploy the site" — and it handles everything.
No other mainstream AI tool does this. ChatGPT's custom instructions are static text. Claude's project knowledge is per-conversation context. Hermes skills are procedural knowledge that compounds.
2. Multi-Surface, Same Agent
This is the feature that surprised me most in daily use.
You start a research task at your desk through the terminal. You step away and continue the same conversation via Telegram on your phone. You come back to your desk and the agent is still there, with full context — because the session and memory are yours, not tied to one interface.
For developers, the most powerful pattern is sub-agent delegation: you tell Hermes in Telegram to run a Claude Code or Codex session on your repo, it spawns that agent as a subprocess, monitors it, and sends you the result when it finishes. You get Claude Code's coding power without sitting at a terminal.
3. Provider Freedom
Most AI tools lock you to one model. This matters more than it seems:
- Claude 4 Sonnet is great at code but expensive for high-volume summarisation
- GPT-5 is strong for creative writing but may not be cheapest for data extraction
- A local model (free) may be sufficient for simple file management tasks
Hermes lets you set a default model, override it per task, and even run different models for different cron jobs — all through the same interface. You pay for exactly the capability each task needs.
Getting Started (5 Minutes)
The quickest path to a working Hermes setup:
# Install
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
# Run the setup wizard (pick your model and provider)
hermes setup
# Start chatting
hermes
You will need an API key from at least one provider. The simplest options are:
| Provider | How to get a key |
|----------|-----------------|
| Nous Portal | Subscribe at portal.nousresearch.com, then hermes portal |
| OpenRouter | Sign up at openrouter.ai, get an API key |
| Anthropic | console.anthropic.com, get API key |
| OpenAI | platform.openai.com, get API key |
To connect Telegram (the most popular messaging surface):
hermes gateway setup
# Follow the prompts to add Telegram bot token
hermes gateway install
hermes gateway start
From that point, you have a persistent AI agent reachable from your phone, with full access to your machine's filesystem, terminal, and tools.
Who Is This For?
Hermes Agent is not the simplest AI tool to set up. The install requires a terminal and at least one API key. So who benefits most from it?
Software developers get the most obvious value: persistent project context, sub-agent delegation to Claude Code/Codex, scheduled CI watchers, and messaging-platform access that means you never have to sit at your desk to monitor a build.
Small business owners and freelancers benefit from the automation layer — scheduled research, cross-platform content workflows, CRM integrations, and the ability to interact with their AI agent from WhatsApp or Telegram instead of a laptop.
Power users of any kind who have outgrown "ask ChatGPT a question" and want an AI that accumulates knowledge about their environment, preferences and workflows over months and years.
If you only need to ask occasional questions, ChatGPT's web interface is still the lowest-friction option. Hermes pays off when AI becomes a daily part of your workflow and you want it to feel like a colleague rather than a search engine.
Limitations to Know
No tool is perfect, and it is worth flagging what Hermes does not do well yet:
- Setup complexity. Compared to signing into a website, installing Hermes requires terminal comfort. The desktop app improves this, but the gateway and messaging setup still involves API keys and configuration files.
- Windows quirks. The core works on Windows, but some POSIX-oriented scripts and tooling assumptions need workarounds (mainly around line endings, keybindings and path conventions).
- Cost management. When your AI agent has full tool access, a runaway loop can generate unexpected API costs. Hermes has safety measures (turn limits, cost display, approval prompts) but monitoring is your responsibility.
- No official mobile app. Mobile access is through messaging platforms (Telegram, Discord, etc.) rather than a dedicated app. For most users this is actually preferable, but it is worth knowing.
Conclusion
The AI tools landscape in 2026 roughly looks like this:
- ChatGPT is your general-purpose assistant — convenient, familiar, limited.
- Claude Code is the best coding agent for Anthropic models — powerful but narrow.
- Codex is OpenAI's answer to Claude Code — excellent within the OpenAI ecosystem.
- Hermes Agent is the open platform that ties everything together: any model, any interface, persistent memory, and automation that compounds over time.
If you have ever wished your AI assistant remembered what you were working on last week, could be reached from your phone, or could switch models depending on the task — Hermes is the tool built for that exact frustration.
It is free, open source, and the setup takes about five minutes. The skills and memory it accumulates over your first month of use are what make it hard to leave.
Extended Reading
- Hermes Agent official documentation
- Hermes Agent on GitHub
- Nous Research — Hermes AI Agents overview
- Nous Portal subscription — managed API access for Hermes
- Related: What Is an AI Agent? — a conceptual introduction to AI agents for non-developers
