How to Set Up and Launch Your Hermes Agent

How to Set Up and Launch Your Hermes Agent — The AI That Gets Smarter Every Day

May 18, 2026

Hermes Agent setup guide — mythic AI wizard at a terminal, starlight and code

⚡ AI Engineering

Hermes is the self-improving AI agent from Nous Research that compounds knowledge across sessions, auto-generates skills from experience, and runs on pennies a day. Here's how to deploy it, what to have it do, and how it compares to OpenClaw.

By Chief Wizard & James Tipton | Category: AI Engineering

May 2026 Determination Development Hermes Agent · AI Automation · Agentic AI

What Is Hermes — And Why Does It Matter

Most AI agents are stateless. Every session starts from zero. You re-explain your business, your tone, your goals — every single time. You curate memory files manually. You write skills by hand. The agent is only as good as the last thing you told it.

Hermes breaks that pattern. Built by Nous Research — the lab behind the Hermes model family — it's an open-source agent with a closed learning loop. After every complex task it auto-generates a reusable Markdown skill. It stores every session in a searchable SQLite database. It builds a persistent model of you, your projects, and your patterns that deepens over time. The longer it runs, the sharper it gets — automatically, without you curating anything.

That compounding is the real product. Not the chat interface. Not the tools. The fact that the agent you have in six months is materially better at your specific workflows than the one you launched today.

Clone your AI agent into a cheaper body without losing the soul. Same voice, same mission, fraction of the cost.

Hermes vs OpenClaw — The Honest Comparison

If you've been running OpenClaw, you already know what a capable agent looks like. Hermes isn't a replacement — it's a different architecture optimized for different things. Here's how they actually compare:

Capability	OpenClaw	Hermes
Skill creation	Manual — you write them	Automatic after complex tasks
Cross-session memory	Flat MEMORY.md, manual curation	SQLite with full-text search + LLM summarization
User modeling	Static files you update	Honcho dialectic modeling — deepens automatically
Provider flexibility	Multi-profile, complex auth routing	OpenRouter-native — one key, 200+ models
Ecosystem / integrations	Massive skill ecosystem, ClawHub	89 bundled skills, growing fast
Platform reach	15+ messaging platforms	15+ messaging platforms
Cost on Haiku	$6–20/day on Sonnet	Pennies per day on Haiku
Migration from OpenClaw	—	Built-in: hermes claw migrate
Provider switching	Complex — layered auth profiles	hermes model — no code changes
Best for	Broad integrations, reach, orchestration	Deep execution, learning, cost efficiency

The hybrid approach: Run OpenClaw as your gateway and broad integration layer. Run Hermes as your deep execution engine. They can share the same workspace and soul files — Chief Wizard stays Chief Wizard across both.

The Smart Workflow: Claude.ai Projects + Hermes for Background Automation

Before we get into Hermes setup, here's the broader workflow that makes this whole system efficient — and cheap.

Claude.ai Projects do most of the heavy thinking. A Project is a persistent workspace where Claude maintains context about your business, your voice, your goals, and your ongoing work across every conversation. You set it up once with your brand guidelines, your site structure, your product roadmap — and every chat inside that project inherits all of it. No re-explaining. No pasting context every session.

Use Claude.ai Projects For

Strategy sessions — product planning, content calendars, launch sequencing
Long-form writing — blog posts, book chapters, sales copy, email sequences
Research and analysis — market research, competitor audits, trend reports
Prompt engineering — building and refining your agent's skill files and SOUL.md
Anything requiring deep reasoning — Opus-tier thinking without Opus-tier automation costs

Use Hermes For

Scheduled background tasks — cron jobs, daily reports, heartbeat checks
Automated posting — X/Twitter, Telegram groups, social pipelines
Engagement automation — follow/unfollow scripts, DM sequences
Customer interactions — email skill responses, live chat support flows
Monitoring bots — Polymarket tracking, price alerts, portfolio summaries

The cost logic: Claude.ai Pro handles your high-reasoning work inside projects for a flat monthly fee. Hermes on Haiku handles your background automation for pennies per action. You're not burning Opus credits on a cron job. You're not paying for a Sonnet response to send a scheduled tweet.

What You Can Have Hermes Do

Once your agent is live, the question is what to point it at. Here are the highest-leverage use cases:

Social Media on Autopilot

Hermes can run follow/unfollow scripts on X, schedule and post content, monitor engagement metrics, and surface high-performing post formats as reusable skills. Set up a daily cron job that pulls trending topics in your niche, drafts three post options, and sends them to your Telegram for one-tap approval before posting. Your audience grows while you sleep.

x-post.skill

Drafts and posts X content in your brand voice on a scheduled cadence.

follow-unfollow.skill

Runs targeted follow/unfollow cycles to grow relevant audience in your niche.

engagement-report.skill

Daily summary of top posts, follower growth, and engagement patterns.

telegram-broadcast.skill

Sends scheduled alpha, updates, and content to your Telegram groups.

Customer Value Through Email and Live Chat

Hermes can monitor your inbox, draft responses to common questions using a skill library you've built up over time, and route complex inquiries to you for review. In a live chat context it can handle first-touch customer interactions, answer product questions, and escalate when needed — all with your voice, your offer stack, and your brand personality loaded from the SOUL.md.

Polymarket and Trading Bot Monitoring

If you're running a prediction market strategy or monitoring a trading bot, Hermes can pull position summaries, alert you to significant market movements, generate daily P&L reports, and surface high-confidence opportunities in your target markets. Pair it with Quant Saber's CFO role and you have a financial intelligence layer running 24/7 without you touching a dashboard.

Content Pipeline Automation

The workflow that unlocks real scale: Hermes monitors trending topics in your niche, drafts initial outlines, generates first drafts, formats them to your CMS spec, and queues them for your review. You handle the creative direction and final polish inside Claude.ai Projects. Hermes handles the volume. Two sites, consistent publishing cadence, minimal manual effort.

The compounding advantage: Every time Hermes successfully completes a complex task — a well-performing post format, a solved technical problem, a customer response that landed well — it can generate a reusable skill from that experience. Over months, your agent gets measurably better at your specific workflows without you doing anything extra.

Here's the full setup sequence. This assumes you're running on Linux or macOS (including a VM). Windows users should run this inside WSL2.

Step 1 — Install Hermes

One command. The installer handles Python, Node.js, ripgrep, ffmpeg, and Playwright automatically. Git is the only prerequisite.

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

VM users: If you run this via a shell exec context (like prlctl exec on Parallels), the installer may land under root's home directory instead of your user. After install, verify ownership with ls -la ~/.hermes and chown if needed: chown -R youruser:youruser ~/.hermes

Step 2 — Create Your Telegram Bot

Open Telegram and message @BotFather. Send /newbot, give your bot a name and username, and save the token it returns. This is your TELEGRAM_BOT_TOKEN. Create a dedicated Telegram group, add the bot, and get the chat ID — you'll need it for group access.

Get your Telegram user ID: Message @userinfobot on Telegram. It returns your numeric user ID. Add this to TELEGRAM_ALLOWED_USERS so only you can interact with the agent.

Step 3 — Get Your OpenRouter API Key

Go to openrouter.ai, create an account, and add credits. Create a dedicated API key for this agent — separate keys per agent makes cost tracking clean. The key format is sk-or-v1-.... Set a spending cap on the key itself under OpenRouter's key settings as a safety net.

Step 4 — Write Your .env File

Create or edit ~/.hermes/.env with these values:

OPENROUTER_API_KEY=sk-or-v1-your-key-here
TELEGRAM_BOT_TOKEN=your-bot-token-here
TELEGRAM_ALLOWED_USERS=your-telegram-user-id
TERMINAL_TIMEOUT=60
TERMINAL_LIFETIME_SECONDS=300

Step 5 — Write Your SOUL.md

This is the most important file. Everything else is plumbing — this is who your agent is. Write it as if you're describing a real person: identity, mission, tone, responsibilities, decision rules, and boundaries. Include two cheat cards: a Cost Card (which model to use for which task) and a Voice Card (tone rules, what to avoid, what to always do). Keep it tight — Haiku has a smaller context window than Opus and a bloated SOUL.md will cost you tokens on every message.

# ~/.hermes/SOUL.md
# Chief Wizard — CTO Agent

You are Chief Wizard, the custom AI CTO built by James Tipton...

## Voice Card
- Slightly theatrical, grounded in competence
- Witty when useful, warm underneath
- Direct about risks and tradeoffs
- Allergic to corporate sludge

## Cost Card
- Default tasks: Haiku (fast, cheap)
- Strategy / deep reasoning: invoke Opus manually with /model
- Background / cron: Haiku always

Step 6 — Fix the Model String in config.yaml

This tripped us up during setup. The model string format matters exactly. Open ~/.hermes/config.yaml and set:

model:
  provider: openrouter
  default: "anthropic/claude-haiku-4-5"

fallback_model:
  provider: openrouter
  model: "openai/gpt-4o-mini"

Common mistake: Using claude-haiku-4-5 instead of anthropic/claude-haiku-4-5. The OpenRouter prefix is required. Without it, the model string fails silently and your fallback fires on every request — you won't notice until you check your costs and see GPT-4o charges instead of Haiku.

Step 7 — Add the Telegram Gateway Config

Add the gateway section to ~/.hermes/config.yaml:

gateway:
  telegram:
    enabled: true
    allowed_chat_ids:
      - your-group-chat-id-here

Step 8 — Start the Gateway

hermes gateway run

You should see the gateway connect to Telegram, the bot come online in your group, and the heartbeat start. Send a test message. If you get a response, you're live.

Start fresh sessions with /new when needed. Haiku has a smaller context window than larger models. If conversations get long, the agent starts losing context and response quality drops. /new starts a clean session without losing your SOUL.md or skills.

Step 9 — Set Up as a Background Service (Optional but Recommended)

If you're running on a Linux VM, register Hermes as a systemd user service so it survives reboots and restarts automatically:

systemctl --user enable hermes-gateway
systemctl --user start hermes-gateway
systemctl --user status hermes-gateway

Key Lessons From Our Setup

Separate API key per agent — one key per agent makes cost tracking clean and lets you cap spend independently
Model string format is exact — anthropic/claude-haiku-4-5 not claude-haiku-4-5 — verify which model is actually responding
System prompt does the heavy lifting — keep SOUL.md tight with cheat cards, not full documentation
Haiku context window is smaller — start fresh sessions with /new for long conversations
One provider, one key — OpenRouter gives you access to every major model through a single API key. Stop switching providers and start routing intelligently
If coming from OpenClaw — run hermes claw migrate --dry-run first to preview what migrates before committing

The Cost Win — By the Numbers

This is where Hermes changes the math on running a persistent AI agent:

Setup	Model	Daily Cost (heavy use)	Monthly Est.
OpenClaw default	Claude Sonnet	$6–20/day	$180–600
Hermes on Haiku	Claude Haiku	$0.05–0.50/day	$1.50–15
Claude.ai Pro	All models	Flat rate	$20/month
Recommended stack	Pro + Haiku agent	—	~$25–35/month total

The same Chief Wizard voice, the same mission, the same content quality — at a fraction of the cost. The savings compound directly into your asset strategy.

Ready to Launch Your Agent?

Read the OpenClaw setup guide first — if you haven't already, the OpenClaw guide covers the foundational concepts that Hermes builds on.
Read the OpenClaw Guide →

Install Hermes and follow this guide — the curl install takes 2 minutes. The SOUL.md is where you'll spend real time. Get that right first.

Join Tech Temple on Telegram — we share build logs, config updates, and real results from running this stack daily.
Join Tech Temple →

Join the Tech Temple on Telegram

Daily alpha, system builds, and real conversations about what AI is actually doing — and earning. This is where the build logs live.

Join Tech Temple →

Keywords

Hermes Agent AI Agent Setup OpenRouter Nous Research AI Automation Agentic AI OpenClaw Alternative Claude Haiku Telegram Bot Self-Improving AI Chief Wizard Determination Development Tech Temple

What Do You Think?

Running Hermes already? Hit a wall during setup? Drop a comment below — questions, configs, and real results welcome. This is where the build conversation happens.

Chief Wizard

Chief Wizard is the custom AI James built to deliver deep research, strategic insight, and transformational transmissions in service of human growth.

James Tipton

James Tipton is the creator of Determination Development, empowering creators with new technology workflows.