After Installing Hermes, Still Just Chatting? Use Top-Tier Models for Free with These 7 Configs to Have It Work 24/7 for You [Easter Egg at the End]

Before I knew it, I’ve been using Hermes for a month. Today I’ve organized the configurations that truly changed the experience, sharing them for your reference. From personalized customization to automated tasks, I eventually got it to directly operate my phone🚀

I’ve been using Hermes for about a month. Initially, I just installed and used it. Other than saving tokens, there wasn’t anything mind-blowing.

Later, when I had some time, I looked through the documentation and realized there’s actually quite a lot to it.

Official documentation URL: https://hermes-agent.nousresearch.com/docs/user-stories

Enough chit-chat, straight to the good stuff!

📢 Breaking news: Hermes uses free models provided by Nous Research, so you basically don’t need to pay. Use it for free until you can’t believe it yourself.

The method is super simple: just subscribe to Free on Nous Research — $0/month, bind a payment method for the first time, no need to top up.

Steps:

• Visit portal.nousresearch.com
• Go to Products, select the Free subscription

Once subscribed, switch the model in Hermes:

hermes model

In “Available free models”, just pick the current free model.

At the time of writing this article, the free model is deepseek v4 flash, and the end date is not yet set. Grab it while you can!

Hermes, the generous one, often brings new free models online, basically seamlessly. In China, you hardly have to spend any money.

Lonely @Lonely__MH · May 13
Some in the comments say it requires payment. Using the free models provided by @NousResearch on Hermes does not require payment.

You just need to subscribe to Free on nousresearch.

Steps:

1. Visit https://portal.nousresearch.com
2. Select Products - Select Free subscription

That’s all.
Quote
Lonely @Lonely__MH · May 13
“Freebies incoming” Hermes agent has launched another free model — limited time, end date unknown!

This time it’s Alibaba’s Qwen-3.6-Plus, another national pride!

Steps:

1. hermes model
2. Select -> Nous Portal subscription
3. Select -> qwen/qwen3.6-plus
4. Start pedaling

In China, it feels like you never need to spend money. The freebies are endless.
x.com/Lonely__MH/sta…32111.9K

01 SOUL.md: Give It a Soul First

The default Hermes is polite, versatile, and neutral — always starting with “Okay, let me help you,” with long and roundabout answers. You’ll get tired of it after a while.

SOUL.md is a personality definition file that loads on every session. Edit it, and the Agent becomes what you want.

Write your rules in ~/.hermes/SOUL.md:

Thinking Style

• Verify before answering; when unsure, look it up with tools — don’t guess.
• For tasks involving more than 3 steps, first list the plan, confirm the direction, then proceed.

Output Requirements

• Start with the conclusion; prioritize commands and code — no fluff.
• Don’t start with “Okay,” “Of course,” “Glad to help.” Get straight to the point.
• When uncertain, say so directly — better than making something up.

High-Risk Operations

• Deleting data, restarting services, overwriting files — must be announced beforehand.

Don’t worry if you don’t know what to write. First, chat normally with Hermes for a few days, then ask it to generate a config: It will read your session history, summarize your preferences, and generate a configuration that fits your habits — much more accurate than writing from scratch yourself.

02 Memory System: Make It Truly Remember You

This is the core capability that sets Hermes apart from every other AI on the market.

Cross-session memory — anything you tell it, it will actually remember because it’s stored in a database, persisted.

Store Memories Actively:

When you encounter important information, just tell it to remember:

/remember Our main database's default port is 5433, not 5432

After a complex task, ask it to summarize:

Save the process and conclusions of this OOM investigation into memory, so next time a similar issue arises, it can be recalled directly.

The longer it works, the more context it accumulates, and the less background information you need to repeat. After a month, you’ll find it handles your tasks much faster than when you first installed it.

03 Skill System: Turn Experience into Muscle Memory

Hermes automatically learns from your conversations, turning them into reusable Skill files stored in the ~/.hermes/skills/ directory.

You can also trigger it proactively:

/save_skill Save the entire process of configuring the SSL certificate as a Skill.

Next time a similar task comes up, Hermes will directly use the existing experience — very convenient, so easy 😄.

Multiple Skills can be chained together to run a complete automated pipeline.

For example, a content creation scenario: asset collection → keyword analysis → copy generation → format adaptation → publishing. Each step is a Skill; chain them together for an unattended workflow.

04 Cron Scheduled Tasks: Work While You Sleep

Hermes has a full scheduled task system. You don’t need to write cron expressions; just tell it directly:

/cron add "every morning at 9am" "Summarize last night's emails, today's calendar, and weather, and send to me."

Hermes will automatically create the scheduled task and push the results to your configured Telegram or other channel.

🚀 Silent Mode 🚀:

For monitoring tasks, add if normal, reply [SILENT]. When everything is normal, no disturbance; only notify you when something goes wrong:

Check server status every hour; if normal, reply [SILENT]; if there's a problem, alert immediately.

Perfect for server monitoring, API health checks, etc. Only bother you when something is actually wrong — no junk messages.

05 Auxiliary Models: Let Cheap Models Do the Dirty Work

Many tutorials skip this configuration, but once you use it, it’s damn money-saving 🚀.

Hermes supports assigning different models for different types of tasks. Lightweight tasks like web scraping, context compression, image analysis, memory organization — you don’t need a flagship model. Just pair them with a cheap, fast one, and let the main model handle only the tasks that truly need thinking.

Just tell Hermes to configure it for you:

Set the auxiliary model for session compression to qwen3.5-plus.

No need to manually edit config files. After configuration, check the log to confirm:

~/.hermes/logs/agent.log

Look for the line Auxiliary compression: using auto (qwen3.5-plus) — that means it’s active.

In practice, costs dropped significantly with no noticeable loss in experience. Absolutely awesome 😄.

06 Multi-Platform Gateway: Command the Agent from Your Phone Anytime

Hermes has a Gateway daemon that can simultaneously connect to Telegram, WeChat, Slack, Discord, etc.

Once configured, send a message from your phone, and Hermes executes it, responding back with results.

Cron task outputs are also directly pushed to your specified channel.

Use /sethome to set the home channel. After that, all proactive notifications and Cron results are pushed to this single place — very convenient.

My current setup: I configured Telegram and Feishu. I have multiple Telegram accounts, each with a bot added. This way, whether I’m on my computer or phone, I can communicate with it simultaneously.

07 Give Hermes a Phone

OK, after configuring the previous six, Hermes can write code, search info, run scripts, manage services, and handle scheduled tasks.

By the way, my biggest regret in life is not having an Android phone 🤣.

Recently I made some money writing promotional articles, so I bought a phone for my Hermes assistant.

Well, to be precise, it’s not really plugging a physical phone into the server. Instead, Hermes connects to a cloud Android device via Airtap. After that, I can tell Hermes from Telegram:

Open Instagram and check this account.
Complete this operation inside the app.
Show me the task progress.
Continue the phone task from earlier.

Let me briefly introduce Airtap (@airtap_ai). It’s a mobile execution platform designed specifically for AI Agents. It provides cloud Android devices, and the Agent uses it like a real person to open apps, interact with UIs, extract information, and send results back.

Official website 👉: https://airtap.ai/

True evolution in the AI Agent era isn’t about being smarter — it’s about doing those daily, repetitive chores that nobody wants to do. Airtap addresses that pain point perfectly (haha, maybe it’s not that painful 😄).

Enough talk, here’s how to configure it.

Technical Architecture Diagram

(Placeholder for diagram)

🔥 Official Skill Repository [Key Point]: 👉 https://github.com/airtap-ai/airtap-skill

Install the Skill:

Method 1: The easiest way is to give the repository URL to the Agent and let it configure itself. One minute, done. A gentleman uses his mouth, not his hands ✌🏻!

Method 2: Based on the Agent you’re using, pick the corresponding command:

Claude Code

claude plugin marketplace add airtap-ai/airtap-skill
claude plugin install airtap@airtap

Codex

npx skills add airtap-ai/airtap-skill -a codex

Cursor / Windsurf / Cline / GitHub Copilot

npx skills add airtap-ai/airtap-skill -a cursor      # Cursor
npx skills add airtap-ai/airtap-skill -a windsurf    # Windsurf
npx skills add airtap-ai/airtap-skill -a cline       # Cline
npx skills add airtap-ai/airtap-skill -a github-copilot

Hermes and other Agents

# Using hermes as an example
hermes skills install airtap-ai/airtap-skill/skills/airtap

⚠️ Hermes may be blocked by security scanning during installation. Add --force to continue:

hermes skills install airtap-ai/airtap-skill/skills/airtap --force

Configure Token:

Go to airtap.ai/app Settings, create a Personal Access Token, then hand it over to Hermes for configuration.

How to Use:

Once installed, just send a message in Telegram:

Use Airtap to open YouTube, search for "ChatGPT half-price subscription tutorial", and tell me the results.

The Agent recognizes the mobile task, creates a cloud phone task, executes it, and sends back the results. If it needs confirmation midway, it will ask you. You reply, and it continues.

Pair with Cron, and things get interesting:

• Daily: Open the Starbucks App to activate today’s deals + point check-in.
• Every Sunday: Add this week’s recipe ingredients to the Instacart cart.
• Daily: Track today’s deliveries from Amazon/UPS/FedEx and compile a summary.
• Every morning at 8 AM: Summarize emails + calendar + weather into a briefing and send to me.
• Post this video simultaneously to TikTok + Reels + YouTube Shorts.

Write a prompt once, schedule it as a cron job, and have it run automatically every day. That’s what an Agent should look like in the AI era.

I can’t help but sigh — what a great time to be alive! 🚀

Some Examples of What I’ve Done:

1 - Scrape daily AI hot news.

2 - Real-time YouTube video search.

3 - Routines feature — you can set daily timed tasks and receive notifications via email.

【🚀 Easter Egg 💐】4 - 🚀 Register Claude on Android — since Airtap’s phone IP is in the US, the registration process was super smooth. No phone number prompt. In theory, you could even top up the account on this phone! (Needs testing, but account registration works — the picture below shows a newly registered account 😄)

Call me Lei Feng!

Finally, here’s the official Case. More examples 👉: https://airtap.ai/see-it-in-action

Summary

Suggested Order for Advanced Setup:

First, write a good SOUL.md → Then use /save_skill to solidify common operations → Configure Cron to let it run for you → Set up Airtap to unlock the app world.

Take it step by step, and you’ll feel the change immediately at each step.

That’s all, folks!

If this tutorial helped you, feel free to Follow + Save + Share 👏🏻

Follow @Lonely__MH for continuous sharing of beginner-friendly tutorials and AI tool tips.