Self-Hosted AI

📅 May 16, 2026 · 8 min read

10 Automation Recipes You Can Run With Your AI Agent Right Now

You've set up your AI agent on a $5 VPS. You can chat with it on WeChat or Lark. Now what?

The real power of a self-hosted AI agent isn't the chat — it's the automation. Your agent can run tasks on a schedule without you touching anything. While you sleep, it's reading the news, checking your servers, analysing market trends, and sending you a briefing before your morning coffee.

Hermes Agent has built-in cron job scheduling — no external tools, no Docker cron containers, no Lambda functions. You tell it what to do and how often, and it just works.

Here are 10 automation recipes you can copy-paste and start using today.

---

☀️ Recipe 1: Daily Morning Briefing

Goal: Every morning at 8 AM, get a concise briefing with top tech news, weather, and your schedule.

Setup (takes 2 minutes):

hermes cron create \
  --name "morning-briefing" \
  --schedule "0 8 * * *" \
  --prompt "Morning briefing:
  1. Search for today's top tech/news headlines
  2. Give me a short summary of 3-5 important stories
  3. End with a tip for running my AI agent better"

That's it. Every morning, your agent researches and sends you a clean briefing. No newsletter subscription needed, no data mining, no AI training on your reading habits.

I get mine on WeChat while I'm still in bed. Takes 30 seconds to read, sets the tone for the day.

🖥️ Recipe 2: Server Health Report

Goal: Know if your VPS is running out of disk space, RAM, or showing unusual activity.

Setup:

hermes cron create \
  --name "server-health-check" \
  --schedule "0 */6 * * *" \
  --prompt "Run these checks on the server:
  - 'df -h' — disk usage
  - 'free -h' — memory usage
  - 'uptime' — how long since last reboot
  - 'top -b -n1 | head -5' — top processes
  If any value is critical (disk > 85%, swap > 50%), alert me loudly.
  Otherwise give a 3-line summary."

This one saved me twice already. Once when a log file grew to fill the disk (caught at 88%), and once when a stalled process was eating 90% CPU. Your agent spots problems before they become emergencies.

🔔 Recipe 3: Website Change Detector

Goal: Get notified when a competitor updates their pricing page or a job board posts new openings.

Setup:

hermes cron create \
  --name "track-competitor" \
  --schedule "0 */12 * * *" \
  --prompt "Visit https://example-competitor.com/pricing
  Note any changes to the content. Look for:
  - New pricing tiers
  - New features mentioned
  - Changed copy
  Summarise what's different since last check."

Great for job seekers tracking specific company career pages, or startups keeping an eye on competitors without paying for expensive monitoring tools.

💰 Recipe 4: Crypto / Stock Price Alert

Goal: Get a price check on your portfolio every few hours — only if there's notable movement.

Setup:

hermes cron create \
  --name "crypto-check" \
  --schedule "0 */4 * * *" \
  --prompt "Check current BTC and ETH prices.
  Compare them to yesterday.
  Only message me if there's more than 3% change.
  If stable, stay silent — I don't need noise."

The "only message if significant" pattern is key. Your agent can stay quiet when everything is normal and only bother you when something matters. That's the difference between useful automation and notification spam.

📰 Recipe 5: Hacker News / Reddit Digest

Goal: Get the top 3-5 posts from Hacker News or a subreddit, curated to your interests.

Setup:

hermes cron create \
  --name "hacker-news-digest" \
  --schedule "0 9,18 * * *" \
  --prompt "Fetch https://hacker-news.firebaseio.com/v0/topstories.json
  Read the top 10 stories.
  Pick the 3 most relevant to: self-hosting, AI, Linux, startups.
  Summarise each in 2-3 sentences with a link."

Two runs per day — morning and evening. You get an AI-curated feed that cuts through the noise. No doom-scrolling needed.

🌤️ Recipe 6: Weather + Traffic Before Commute

Goal: Know if you need an umbrella or should leave early for work.

Setup:

hermes cron create \
  --name "commute-forecast" \
  --schedule "30 7 * * 1-5" \
  --prompt "Check today's weather for Singapore.
  Is it going to rain?
  Give me a one-line recommendation:
  '☀️ Clear all day' or '🌧️ Rain from 3pm — bring umbrella'"

Schedule it for weekdays only (1-5 in cron) so it doesn't bother you on weekends.

🧹 Recipe 7: Log File Cleanup & Housekeeping

Goal: Prevent disk space issues by keeping your server tidy.

Setup:

hermes cron create \
  --name "log-cleanup" \
  --schedule "0 3 * * 0" \
  --prompt "Run:
  - 'journalctl --vacuum-time=7d' — keep 7 days of system logs
  - 'find /var/log -name \"*.gz\" -delete' — remove old compressed logs
  - 'docker system prune -af' if Docker is installed
  - 'df -h /' — report freed space
  Send me a summary of what was cleaned up and how much space was reclaimed."

I run this every Sunday at 3 AM. It keeps my $5 VPS lean — I've reclaimed 2-3GB of disk space in a single run. On a 25GB disk, that's huge.

📊 Recipe 8: Weekly Analytics Report

Goal: See how your website or business is performing week-over-week.

Setup:

hermes cron create \
  --name "weekly-report" \
  --schedule "0 10 * * 1" \
  --prompt "Check Gumroad sales for my products this week.
  Check website analytics for trevva.io if available.
  Summarise:
  - Any new sales / visitors
  - Notable changes from last week
  - One suggestion for this week's focus"

Runs Monday morning. You get a quick dashboard without logging into any analytics tool. If you run a side project or small business, this is incredibly useful.

🔄 Recipe 9: Cross-Agent Sync

Goal: One agent runs scheduled tasks and syncs results to another agent on a different machine.

Setup (advanced):

If you run multiple Hermes instances (I have one on my cloud VPS and one on my Windows laptop), you can set them up to talk to each other using the built-in file-based message queue:

# On your main VPS agent
hermes cron create \
  --name "cross-vps-sync" \
  --schedule "0 */12 * * *" \
  --prompt "Check the message queue for pending tasks from the GCP agent.
  Process any tasks found (web research, data analysis).
  Queue responses back."

Your laptop agent can research things, your VPS agent can run heavy tasks, and they coordinate without sharing a bot account. Multi-agent orchestration without the complexity.

🪄 Recipe 10: AI-Powered Idea Generator

Goal: Get a fresh content idea or business concept delivered every morning to kickstart your creativity.

Setup:

hermes cron create \
  --name "idea-generator" \
  --schedule "0 7 * * *" \
  --prompt "Think of one blog post idea or side project concept
  related to: AI agents, self-hosting, or developer tools.
  Make it specific and actionable.
  Include:
  1. The idea in one sentence
  2. Why it would be useful
  3. Rough implementation steps (3 bullets max)"

Writer's block cure. I got the idea for my ResumeReview tool from one of these morning prompts.

---

⚙️ How Hermes Cron Jobs Actually Work

Unlike traditional cron (which runs shell scripts), Hermes cron jobs run through the AI agent itself. When a job triggers, Hermes wakes up, reads your prompt, and uses its tools (web search, file system, terminal commands) to complete the task. Then it delivers the result to your chat.

Key features:

• Natural language prompts — no scripts, no bash. Just tell your agent what you want in plain English.
• Decision intelligence — your agent decides to stay silent if nothing interesting happened. No empty reports.
• Multi-platform delivery — get results on WeChat, Lark, Telegram, or email. Choose per job.
• Skills support — attach reusable skill files to cron jobs for specialised workflows.
• Script mode — for power users, you can attach shell scripts that run before the agent processes the output.
• Per-job model override — use Gemini free tier for simple checks, GPT-4 for complex analysis.

🔑 The Secret Pattern: Think in Conditions, Not Timers

Most people set up cron jobs as "send me X report every Y hours." That's table stakes. The magic is in conditional delivery:

• "Only message me if disk > 85%" — alert, not noise
• "Compare to yesterday and only report significant changes" — trend-aware
• "Skip if nothing changed" — zero clutter
• "If critical, send to both Lark and Telegram" — redundancy for emergencies

Your AI agent has judgement. Use it. A well-designed cron job runs quietly 90% of the time and is worth its weight in gold the other 10%.

---

Real Example: What I Run Daily

Here's my actual cron schedule right now:

Job	Schedule	Delivers	Silent When
Morning reminder	9 AM daily	Lark (origin)	Never — always shows up
Queue poll (Lyx)	Every minute	Local	Always — only processes tasks
Queue poll (GCP)	Every minute	Local	Always — only processes tasks
Blog traffic check	Once a day	Lark	Under threshold

The queue pollers run every single minute but they're completely silent. They check for inter-agent messages and process them. The only time I hear from them is if something goes wrong. That's the ideal pattern.

---

🔧 Getting Started in 2 Minutes

If you already have Hermes Agent running:

# List your current jobs
hermes cron list

# Create your first automation
hermes cron create \
  --name "my-first-job" \
  --schedule "0 9 * * *" \
  --prompt "Tell me something interesting about AI that happened today"

Then watch. Tomorrow at 9 AM, your agent will research and message you. No scripts, no cron syntax learning, no Docker compose files.

Or grab the free PDF from trevva.io/guide and the full source code on GitHub.

💡 New to VPS hosting? DigitalOcean offers $200 free credit for new users — run all these automations free for months.

📅 May 14, 2026 · 7 min read

Open-Source AI Agents You Can Self-Host in 2026

AI agents are exploding in popularity. From personal assistants that manage your email to coding agents that fix bugs autonomously — there's a self-hosted option for almost everything.

But with so many projects hitting 100k+ stars on GitHub, choosing the right one can be overwhelming. Some are great for chat, others for desktop automation, and some for complex multi-agent workflows.

I've been running self-hosted AI agents for months. Here's my honest breakdown of the best open-source options in 2026 — what they're good at, where they fall short, and what they'll actually cost you.

Why Self-Host in 2026?

Before diving into specific tools, let's address the elephant in the room: why not just use ChatGPT, Claude, or a managed service?

• Privacy. Your conversations, files, and context stay on your hardware. No third-party training data.
• Cost. A $5/month VPS + a free Gemini API key runs a 24/7 agent for nearly zero ongoing cost.
• Customizability. Want your agent to send WeChat messages, run cron jobs, or control your smart home? Self-hosted = you decide.
• Always-on. No subscription limits, no downtime from provider outages.

The trade-off? You'll need basic Linux skills and about 30 minutes to an hour of setup time. But once it's running, it runs forever.

---

🐾 Hermes Agent — The 24/7 Personal Assistant

Best for: Message-based AI assistant that runs 24/7 on cheap hardware

GitHub: github.com/NousResearch/hermes-agent

Language: Python

Hermes Agent is what I use daily. It's an open-source AI agent designed to live on a VPS and communicate with you through messaging platforms like WeChat, Lark (Feishu), Telegram, Discord, and WhatsApp.

What makes it unique:

• Persistent memory — remembers you across sessions. It knows your preferences, past conversations, and project context.
• Cron jobs — built-in scheduling. Daily briefings, health checks, price alerts — all without extra config.
• Multi-agent support — can delegate tasks to other Hermes instances or subagents.
• File-based messaging — Hermes instances can communicate via a message queue without relying on a shared bot account.
• Tools & skills system — extensible with custom Python/CLI tools, MCP servers, and plugins.

Cost to run: $5/month for a VPS. AI model can be Gemini free tier ($0).

Setup time: ~30 minutes with the complete guide.

---

🦞 OpenClaw — The Desktop Automation AI

Best for: AI that controls your computer — manages files, browses the web, runs apps

Website: openclaw.ai

GitHub: github.com/openclaw/openclaw (372k ⭐)

Language: TypeScript

OpenClaw is one of the fastest-growing open-source projects of 2026 — and for good reason. It's an AI assistant that lives on your computer, can see your screen, use your apps, and do real work for you.

Think of it as a digital coworker you message on WhatsApp or Telegram: "Clear my inbox, draft replies, and block 3-4pm for focus time." It does it.

What makes it unique:

• Desktop control — can see your screen, click buttons, type text, and navigate apps
• Proactive background tasks — cron jobs, reminders, heartbeat monitoring
• 50+ integrations — WhatsApp, Telegram, Discord, Slack, Gmail, GitHub, and more
• Skills registry — 5,400+ community-built skills for every use case
• Cross-platform — macOS, Windows, Linux, Docker

Cost to run: Free (open source, MIT). Self-host on your laptop or VPS.

Setup time: ~15 minutes with npx openclaw@latest. But running it 24/7 on a VPS requires Docker, networking setup, and ongoing maintenance.

Caveat: OpenClaw is powerful but complex. It requires Node.js, pnpm, and careful configuration for 24/7 operation. Several managed hosting services (like MyClaw.ai) charge $16-133/month to handle this for you.

If you just want an AI assistant you can chat with on WeChat or Lark without the desktop automation complexity, Hermes Agent is simpler to set up and cheaper to run.

---

🧠 Open Interpreter — Code & System Automation

Best for: Running code, analyzing data, and controlling your system via natural language

GitHub: github.com/openinterpreter/open-interpreter

Open Interpreter lets LLMs run code (Python, JavaScript, Shell) locally. It's like ChatGPT's Code Interpreter but open-source and running on your own machine.

Use cases:

• Analyze datasets and create visualizations
• Automate file operations — rename, convert, organize thousands of files
• Control web browsers programmatically
• Install packages and run system commands via chat

Cost: Free. Uses your own API keys.

Limitation: Not designed for 24/7 operation or messaging platform integration. Best used interactively.

---

👥 CrewAI — Multi-Agent Workflows

Best for: Orchestrating multiple AI agents to work together on complex tasks

GitHub: github.com/crewAIInc/crewAI

Language: Python

CrewAI is a framework for orchestrating role-based AI agents. You define agents with specific roles (researcher, writer, reviewer) and they collaborate to complete tasks.

Example workflow:

• Researcher agent — gathers data from the web
• Writer agent — drafts a report based on findings
• Reviewer agent — checks for accuracy and quality

Cost: Free. Uses your own API keys.

Limitation: More of a framework than a ready-to-use assistant. Requires Python coding to set up. Not designed for messaging integration out of the box.

---

🔧 Dify — Low-Code AI Application Platform

Best for: Building AI applications visually without coding

GitHub: github.com/langgenius/dify

Dify is a visual AI application platform with a drag-and-drop workflow builder. You can create chatbots, text generators, and AI workflows without writing code.

Use cases:

• Customer support chatbots with RAG (retrieval-augmented generation)
• Content generation pipelines
• AI-powered data analysis dashboards

Cost: Free (self-hosted) or cloud tier available.

Limitation: Requires Docker and more resources (2GB+ RAM recommended). More of a platform than a personal assistant.

---

Quick Comparison

Tool	Best For	24/7	Messaging	Min Cost	Difficulty
🐾 Hermes Agent	Personal assistant	✅	✅ WeChat, Lark, Telegram	$5/mo	Low
🦞 OpenClaw	Desktop automation	⚠️ (needs VPS)	✅ WhatsApp, Telegram	Free / $16+ hosted	Medium-High
🧠 Open Interpreter	Code execution	❌	❌	Free	Low
👥 CrewAI	Multi-agent workflows	⚠️ (manual setup)	❌	Free	Medium
🔧 Dify	AI app builder	✅ (Docker)	⚠️ (API-based)	Free	Medium

---

Which One Should You Choose?

Choose Hermes Agent if: You want a 24/7 AI assistant you can chat with on WeChat, Lark, or Telegram. You want zero ongoing cost beyond a $5 VPS. You want persistent memory, cron jobs, and easy setup.

Choose OpenClaw if: You need an AI that can control your computer — manage files, browse the web, or run desktop apps. You're comfortable with TypeScript and Docker setup.

Choose Open Interpreter if: You want an AI coding assistant that runs locally and can execute code, analyze data, and control your terminal.

Choose CrewAI if: You're building complex multi-agent workflows and need a framework to orchestrate them.

Choose Dify if: You want to build AI applications visually without coding, especially chatbots with RAG.

---

My Personal Take

I run all my personal AI needs on Hermes Agent — it handles my daily briefings, server monitoring, research queries, and task automation. I access it through Lark on my phone and it costs me exactly $5/month for the VPS. The AI model (Gemini free tier) costs $0.

I'm also experimenting with OpenClaw on my laptop for desktop automation — it's impressive but overkill for the messaging-based assistant use case most people need.

The beauty of open-source AI agents in 2026 is that you're not locked into any one tool. Start with something simple (Hermes Agent takes 30 minutes to set up), and layer in more specialized tools as your needs grow.

Or grab the free PDF from trevva.io/guide and the full source code on GitHub.

💡 Planning to start with a VPS? DigitalOcean offers $200 free credit for new users — enough to run your agent free for months.

📅 May 12, 2026 · 3 min read

Get Your Resume Reviewed by AI — Free

Looking for a job? Your resume needs to pass through AI filters before a human ever sees it. I built a free tool that analyses your resume and gives you actionable feedback — what's missing, what to reword, and how to optimise for ATS (Applicant Tracking Systems).

What It Checks

• ATS compatibility — can automated systems parse your resume?
• Keyword density — are you using the right terms for your industry?
• Formatting issues — tables, columns, and graphics that confuse parsers
• Content gaps — missing sections like summary, certifications, or metrics
• Bullet point quality — weak vs. impact-driven phrasing

Built with FastAPI + OpenAI vision — your resume is analysed privately and never stored.

More features coming soon: side-by-side comparison, cover letter generator, and interview question prep based on your resume.

📅 May 12, 2026 · 4 min read

The Complete Hermes Agent Blueprint — From Zero to 24/7 Automation

I wrote a comprehensive 11-chapter guide on setting up and running Hermes Agent on a $5/month VPS. It covers everything I learned from months of trial and error — so you don't have to make the same mistakes.

What's Inside

• Chapter 1–3: Server setup, Hermes installation, first launch
• Chapter 4–5: WeChat & Telegram integration (QR code, webhook, everything)
• Chapter 6: Configuring AI providers — Gemini free tier, GPT, Claude, DeepSeek
• Chapter 7: Cron jobs — daily briefings, health checks, price alerts
• Chapter 8: Multi-agent workflows — delegate tasks between agents
• Chapter 9: Security hardening — firewall, SSH keys, fail2ban
• Chapter 10: Tools & skills — custom tools, MCP servers, plugins
• Chapter 11: Backup, restore, and migration between VPS providers

Or grab the free PDF from trevva.io/guide.

📅 May 11, 2026 · 5 min read

Gemini Free Tier Is All You Need to Run an AI Agent 24/7

Most people think running an AI agent costs money every month. API tokens, model pay-per-use, hosting fees — it adds up. But there's a secret: Google's Gemini free tier gives you 60 requests per minute completely free. That's more than enough to run a 24/7 personal AI assistant.

I've been using it for months and my total monthly cost for the entire setup is just $5 for the VPS server. Zero API costs.

What Gemini Free Tier Gets You

• Gemini 2.0 Flash — fast, capable model
• 60 requests per minute — plenty for a personal assistant
• 1,000 requests per day comfortably covered
• Free rate-limited tier, no credit card required to start

Compare that to OpenAI: even gpt-4o-mini costs about $0.15 per million input tokens. For a 24/7 agent sending you briefings, answering questions, and running cron jobs, that adds up.

How to Set It Up

You need two things:

• A $5 VPS server (or any Linux server)
• A Google AI Studio API key (free, get it at aistudio.google.com/apikey)

Once you have those:

pip install hermes-agent
hermes setup

During the setup wizard, choose Google/Gemini as your provider and paste your API key. That's it.

What I Run on the Free Tier

With zero API cost, my agent runs these every day:

• ☀️ 8 AM briefing — news, weather, schedule summary
• 📊 Server health check — disk, RAM, uptime every 30 min
• 💰 Crypto check — BTC/ETH prices every 4 hours
• 💬 On-demand chat — ask questions, get research, write code

None of this costs a cent beyond the server itself.

---

Limitations (Honest Take)

The Gemini free tier is not perfect:

• Rate limits can be hit if you run many concurrent tasks
• Not as powerful as GPT-4 or Claude for complex reasoning
• No vision in the free tier (Gemini Pro Vision costs)

But for 95% of personal AI assistant use — briefings, reminders, questions, code help — it's more than enough. And the price is right: $0.

Tip: You can also configure fallback providers in Hermes. Use Gemini free as your default, and switch to a paid model only for complex tasks.

📅 May 11, 2026 · 8 min read

How to Run an AI Agent 24/7 on a $5/Month VPS

Want your own AI assistant that runs 24/7, connects to WeChat or Telegram, remembers everything you tell it, and costs less than a cup of coffee per month? It's easier than you think.

What You'll Need

• A VPS server (I use DigitalOcean or Tencent Cloud Lighthouse — both start at $5/month)
• An AI provider API key (Gemini free tier works great)
• About 30 minutes

Step 1: Get a Server

For this guide I'm using a $5 VPS. Both DigitalOcean (get $200 free credit) and Tencent Cloud Lighthouse work perfectly.

I recommend the following specs for running an AI agent:

• 1 vCPU, 1GB RAM (enough for Hermes Agent)
• 25GB SSD
• Ubuntu 22.04 or 24.04 LTS

Step 2: Install Hermes Agent

Hermes Agent is an open-source AI agent that runs on your own hardware. It connects to messaging platforms, has permanent memory, and can run automated tasks on a schedule.

pip install hermes-agent
hermes setup

Step 3: Connect It to WeChat or Telegram

Hermes supports WeChat, Telegram, Discord, Feishu/Lark, WhatsApp, and more. During hermes setup, it'll walk you through generating API tokens and QR code scanning for WeChat.

Step 4: Automate Everything

• ☀️ 8 AM daily news briefing
• 📊 Server health check every 30 minutes
• 💰 Crypto price alerts
• 📡 RSS feed summaries

Or grab the free PDF on trevva.io/guide.

---

Recommended VPS Providers

• DigitalOcean — $5/month, $200 free credit for new users
• Tencent Cloud Lighthouse — $5/month, good for Asia/Singapore users