Hermes Agent: The Complete Guide — From Zero to Self-Improving AI Employee

What this guide covers

This is a soup-to-nuts guide to running Hermes Agent as a 24/7 autonomous "AI employee" — from a single install command to a self-improving multi-agent setup. It walks through ten layers: what Hermes is, how it compares to alternatives, installation, model selection, messaging, first-day setup, the dashboard, use cases, the self-improvement loop, and security.

Bookmark this — you'll need it when you start building.

Layer 1 — What Hermes Agent actually is

Hermes Agent is a 24/7 autonomous AI employee built by Nous Research. It works while you sleep, proactively figures out tasks aligned with your goals, and gets better every session.

Three things separate it from everything else:

• Memory. Everything lives in markdown files on your computer — not the cloud, not a black box. You can read it, edit it, delete it. Full transparency.
• Self-improvement. Every task it completes, it reviews: what worked, what didn't, how to do it better. It edits its own skills after every session.
• Session recall. Every conversation is logged with FTS5 full-text search and LLM summarization. Ask what you talked about three months ago — it knows.

Layer 2 — Hermes vs other tools

Three tools, three different jobs. Here's where each one fits.

Hermes vs OpenClaw. The author's take: OpenClaw has gotten bloated and slow, and updates tend to break setups. Hermes is lighter, snappier, and updates don't destroy your configuration — that reliability is the main reason to switch.

Additional Hermes advantages cited:

• Built-in multi-agent via Kanban (v0.12.0+) — agents claim tasks from a board, work in parallel, hand off when blocked
• Nous Portal with curated models built in
• 166 tracked skills (87 bundled + 79 optional) across 26+ categories
• 20+ messaging platforms (Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Teams, and more)

Hermes vs Claude Code / Codex. Different jobs — use both:

• Hermes = your general-purpose employee. Day-to-day tasks, research, documents, spreadsheets, computer administration, business advice, prototypes — anything that should improve over time. Think Chief of Staff.
• Claude Code / Codex = deep, focused coding sessions. Large complex apps, end-to-end testing, locked-in heads-down work.

Layer 3 — Installation

One command.

Linux / macOS / WSL2:

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

Windows (native PowerShell, early beta):

iex (irm https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.ps1)

Android (Termux): same curl one-liner as Linux — the installer auto-detects Termux.

After install, start with:

hermes

Quick setup walks you through model selection and messaging platform. If you have OpenClaw installed, you'll get an option to import memories — the author recommends starting clean, since two separate agents with separate memories and skills beats merging everything.

Layer 4 — Model selection

Three tiers. The right choice depends on the work, not just budget.

• Expensive — claude-opus-4 / claude-sonnet-4. Best for complex reasoning, long /goal tasks, nuanced writing, and the business-advisor role. Note: Anthropic disabled OAuth for agents, so an API key is required (pay per token).
• Moderate — GPT-5.5. Best for coding, prototyping, and a budget-conscious daily driver. Works with an existing ChatGPT subscription. A good starting point if you're new.
• Affordable — Qwen 3.7 Max, Grok, Nous Portal. Qwen 3.7 Max excels at long-horizon autonomous tasks (35 hours continuous, 1,000+ tool calls). Grok is strong if you already pay for SuperGrok and works for X tasks. Nous Portal is $20/month flat with curated models and no API-key management.

Switch models anytime — no code changes, no reinstall. Different profiles can run different models simultaneously:

hermes model

Layer 5 — Messaging platform

The recommendation: Telegram. It's the only messaging platform actively building for AI agents (topics, agent-to-agent communication, constant new features) and it's free. Setup takes about five minutes — Hermes walks you through copying a token from Telegram's BotFather.

Other supported platforms if you need them: Discord, Slack, WhatsApp, Signal, Matrix, Mattermost, Email, SMS, DingTalk, Feishu, Microsoft Teams, Google Chat, and more — 20+ total from one gateway process.

Layer 6 — First things to do

Step 1 — Tell Hermes about yourself. Send a first message covering your name, what you do, what you're building, your goals for the next 3–6 months, and how you work. This goes into memory, and every proactive task is filtered through it.

Step 2 — Set up your first cron job. Cron jobs are scheduled autonomous tasks described in plain English. For example, ask it to build a small useful micro-app, UI, or system every night at 2am that moves you toward your goals — and wake up to something new each morning.

Step 3 — Learn /goal. This is the most powerful command in Hermes. It turns the agent from a reactive chatbot into a background worker: you set an objective, it breaks it into tasks and executes until done.

/goal [description]     # start autonomous execution
/goal status            # check what's running
/goal pause             # pause without losing context
/goal resume            # continue after pause
/goal clear             # end the current goal
/subgoal [text]         # add conditions mid-execution

Layer 7 — The dashboard (most people use it wrong)

hermes dashboard

Opens in your browser at localhost:9119. The author's advice: open the Skills tab first — that's where the real value is.

• Models tab — swap models instantly, set different models per profile.
• Cron tab — see all scheduled tasks and build complex ones with more control.
• Skills tab — browse, toggle, and read every learned skill. A well-used agent has 150+ skills. Turn on Browser automation, Computer use, Image generation, and Video generation immediately.
• Plugins tab — extra capabilities via API keys (browser-use, fire-crawl, computer-use).
• Profiles tab — multi-agent setup. One profile = one agent with its own memories, skills, and model. Run several specialized roles simultaneously.

Kanban board — the most powerful screen. Each morning, drop every AI-handleable task into Triage and walk away. Hermes splits each task into subtasks, moves them to To-Do, assigns sub-agents, and they execute in parallel.

Statuses: Triage → To-Do → Ready → In Progress → Blocked → Done. The daemon runs continuously (v0.16+), checking for new tasks every 60 seconds — no cron-based polling, no wasted tokens between tasks.

Layer 8 — Use cases

1. Daily tutor — paste a YouTube link; Hermes pulls the transcript, extracts key concepts, and schedules a morning lesson + quiz.
2. Computer administrator — with Tailscale on all your devices, move any file between machines from your phone, anywhere.
3. Session recall — ask it to recall every business idea or link from last month; FTS5 search + summarization spans your whole history.
4. X content workflow with xurl — combine the xurl skill with /goal, research skills, and memory into a recurring content system (data collection → style check → repetition check → draft → quality score → publish). Don't auto-publish on day one — review 5–7 runs first.
5. Mission control — ask Hermes to build a custom dashboard (content pipeline, memory wiki, artifacts page) with no code.
6. Prototype builder — describe a landing page from your phone; it uses your known stack and deploys to localhost.
7. Business advisor — it knows your business, goals, and constraints, so advice is grounded in your actual situation.
8. Overnight /goal runs — hand it a complex task before bed (e.g., a competitor research report) and wake up to the finished doc.

For complex overnight tasks, raise max_turns only when genuinely needed (every turn costs tokens):

hermes config set goals.max_turns 20    # research, reports, content drafts
hermes config set goals.max_turns 50    # code refactoring, multi-step builds

9. Multi-agent org chart — create separate profiles (Chief of Staff, Head of Research, Head of Content), each with its own soul.md, running in parallel and reporting into one morning brief.

Layer 9 — Self-improvement (the actual edge)

The self-improvement loop is what the author considers Hermes's real differentiator:

1. You give Hermes a task
2. It executes
3. After completion it reviews what worked, what didn't, and the optimal path
4. It saves that as a skill in ~/.hermes/skills/
5. Next time, it uses that skill directly

Correct it once, and it doesn't repeat the mistake. Skills are transparent markdown files you can open, read, and edit. Updates from Nous Research add new skills automatically without breaking existing ones.

hermes tools

Turn on immediately: browser-automation, computer-use.

Layer 10 — Security (the honest take)

The author's view: security concerns are overrated for basic personal use, because the agent only does what you tell it to. The main risk is instructing it to do something catastrophic — so think before you prompt and review destructive actions.

For personal use you mostly need common sense, prompt review, and a ground rule in your soul.md such as "Never send money to anyone without explicit confirmation." If something breaks, open the Hermes folder in Claude Code or Codex and ask it to fix the problem.

For production agents that touch sensitive systems, Hermes ships a proper security stack:

Layer 1 — Bitwarden Secrets Manager (credential management):

hermes secrets bitwarden setup   # wizard: installs bws, prompts for token
hermes secrets bitwarden status  # verify connection
hermes secrets bitwarden sync    # dry-run: see what gets applied

One bootstrap token lives in .env; all real credentials live in Bitwarden. Rotate a key once in the web app and every instance picks it up on next restart.

Layer 2 — iron-proxy egress firewall (credential protection):

hermes egress install   # downloads iron-proxy binary, SHA-256 verified
hermes egress setup     # interactive wizard
hermes egress start     # spawn managed proxy daemon
hermes egress status    # binary + config + active mappings
hermes egress setup --from-bitwarden  # pull real credentials from BSM at proxy startup

Instead of injecting real credentials into the sandbox, Hermes hands the agent opaque proxy tokens; iron-proxy swaps them for the real credential at the network boundary. Compromise the sandbox and the attacker only gets tokens that work from behind the proxy. The two layers compose: rotate in Bitwarden, and it propagates across the fleet automatically.

The real insight

ChatGPT and Claude are powerful, but each conversation starts from zero — no memory, no improvement, no context. Hermes compounds:

• Day 1: it knows nothing about you
• Month 1: it knows how you work and what you're building
• Month 6: it knows how you think, your daily tasks, and the optimal way to do each one

The author's framing: memory, the improvement loop, and trust are the real bottlenecks in AI agents — and Hermes addresses all three. The agent itself is open source and $0.

Resources

Official:

• Hermes Agent Docs — installation, configuration, full CLI reference
• Skills Hub — community skills to browse and install
• GitHub — source, issues, PRs

This guide was written and shared by YanXbt, who also publishes companion deep-dives on the /goal playbook, the full /goal guide, the xurl content system, and the Hermes + Bitwarden security stack.