Build a team of self-running digital employees with the Hermes Agent, powered by Xiaomi MIMO (or any LLM)
"AI shouldn't just be a tool you query. It should be a digital employee that runs on its own."
English · 简体中文
Demo · Quick Start · Templates · How It Works
A collection of automation templates built on top of Hermes Agent that help you spin up a daily, hands-off push system in minutes.
I run 10+ scheduled jobs with this setup. Every day it automatically delivers:
- 📰 AI news digest (scrapes Hacker News + tech sites + Twitter)
- 📚 Tech learning (a 40-day course, advanced day by day)
- 🎧 English listening (auto-generated TTS audio)
- 🎯 Interview questions (C++ / CUDA / system design)
- 💰 Personal finance (a 30-day starter plan)
- 🤔 Daily thinking prompts (rotating across 7 domains)
- 🤖 A nightly self-review where the agent "dreams" — recaps yesterday and updates its own memory
All unattended. Jobs fire in the early morning, and the results are waiting in Telegram when you wake up.
Everything runs at night — by morning your Telegram is already lined up for you.
- Install Hermes Agent
- Configure an LLM provider (MIMO / DeepSeek / OpenAI / Claude all work)
- Configure Telegram or another messaging platform
# 1. Clone the repo
git clone https://github.com/schbxg/hermes-daily-automation.git
cd hermes-daily-automation
# 2. Copy the config template
cp config-template.yaml ~/.hermes/config.yaml
# 3. Create a scheduled job (AI news as an example)
hermes cron create "0 3 * * *" --name "AI News" --prompt "$(cat cron-jobs/ai-news.md)"
# 4. List your jobs
hermes cron list# Edit the prompt template
vim cron-jobs/ai-news.md
# Create the scheduled job
hermes cron create "0 8 * * *" --name "My Daily" --prompt "$(cat cron-jobs/ai-news.md)"
# Test run it
hermes cron run <job_id>| Template | Description | Suggested Schedule |
|---|---|---|
| AI News | Scrapes HN + tech sites + Twitter, generates a digest | Daily 3:00 |
| Tweet Drafts | Generates 2-3 tweet drafts daily for human review | Daily 3:00 |
| Template | Description | Suggested Schedule |
|---|---|---|
| Tech Learning | Advances through a tech course on a plan | Daily 5:00 |
| English | Vocabulary + listening + TTS audio | Daily 7:30 |
| Interview Prep | One interview question + answer per day | Daily 7:00 |
| Code Reading | Read a huge file over N days, then auto-pauses itself | Daily 20:00 |
| Topic Study | N-day deep dive on one topic, with real-code examples | Daily 21:00 |
| Template | Description | Suggested Schedule |
|---|---|---|
| Daily Review ("Dreaming") | At 3 AM the agent reviews yesterday's sessions, writes lessons to memory, and suggests new skills | Daily 3:00 |
| Template | Description | Suggested Schedule |
|---|---|---|
| Daily Thinking | Rotates through philosophy / creativity / decision-making | Daily 8:00 |
| Finance | A 30-day personal finance starter plan | Daily 21:00 |
Reusable, parameterized prompt building blocks you can adapt for your own jobs — see prompt-templates/.
| File | Description |
|---|---|
| AGENTS template | Project-level AI onboarding file template |
| Content Progression | A Skill that prevents cron jobs from repeating content |
┌─────────────────────────────────────────────┐
│ Hermes Agent (framework) │
│ ┌─────────┐ ┌─────────┐ ┌─────────────┐ │
│ │ Cron │ │ Memory │ │ Skills │ │
│ │ schedule│ │ system │ │ experience │ │
│ └────┬────┘ └────┬────┘ └──────┬──────┘ │
│ │ │ │ │
│ ┌────▼───────────▼─────────────▼──────┐ │
│ │ Orchestration Loop │ │
│ │ (Prompt → LLM → Tool) │ │
│ └────┬────────────────────────────────┘ │
│ │ │
└───────┼────────────────────────────────────┘
│
▼
┌───────────────┐
│ LLM Model │ ← MIMO / DeepSeek / Claude / GPT
└───────┬───────┘
│
▼
┌───────────────┐
│ Messaging │ ← Telegram / Discord / Slack
└───────────────┘
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Plan file │────▶│ Cron Job │────▶│ History dir │
│ (static) │ │ (daily fire) │ │ (accumulates)│
└──────────────┘ └──────┬───────┘ └──────────────┘
│
1. Read plan → what day is it?
2. Read history → what was covered?
3. Generate content → no repeats
4. Save to history dir
The key idea: use the file system as the agent's external memory so continuity holds across sessions.
Not every job should run forever. A course or a code-reading plan has an end. Each such job
computes "what day is it", and once the plan is over it calls
cronjob(action='pause', job_id=<self>) to pause itself — no zombie jobs, no manual cleanup.
Day 1 ─▶ Day 2 ─▶ ... ─▶ Day N ─▶ "plan finished" ─▶ pause(self)
Learning jobs don't just explain concepts in the abstract. You point them at your project directory, and they pull real examples from your actual code (with file references). What you learn maps directly onto what you're building, so it sticks.
# ❌ Bad: 7 jobs all at 3:00
3:00 - job1, job2, job3, job4, job5, job6, job7
# ✅ Good: spread across the day
3:00 - AI News
5:00 - Tech Learning
7:00 - Interview Prep
7:30 - English
8:00 - Daily Thinking
21:00 - Finance + Tweet reminders## Step 1: Determine what day it is
Run `date +%Y-%m-%d`, compute days elapsed since Day 1.
## Step 2: Read history
ls /path/to/history/ # read the last 3 days
## Step 3: Generate content
Make sure it differs from history.
## Step 4: Save to history
Write to /path/to/history/day_XX.md# Project overview
# Directory structure
# Code conventions
# Known pitfalls
# WorkflowNo. Every template combines:
- A plan file (static framework)
- A history directory (accumulating record)
- A "what day is it" anchor (time anchor)
Together these guarantee content progresses without repeating.
| Model | Best For | Cost |
|---|---|---|
| Xiaomi MIMO | Everyday tasks, Chinese content | Cheap |
| DeepSeek | Tech learning, code analysis | Very cheap |
| Claude Sonnet | Complex reasoning, long docs | Medium |
| GPT-4o | General-purpose | Medium |
Recommendation: MIMO/DeepSeek for daily tasks, Claude/GPT for complex ones.
Yes. Hermes supports 10+ messaging platforms with the same config approach.
# Trigger once manually
hermes cron run <job_id>
# Follow logs
hermes logs --follow
# Pause a job
hermes cron pause <job_id>PRs adding new cron job templates are welcome!
Template format:
# Job name
[Describe the job's purpose]
## Steps
1. ...
2. ...
## Output format
[Define the output template]- Hermes Agent — the AI agent framework
- Nous Research — the team behind the framework
- Xiaomi MIMO — LLM support
If you find this useful, please drop a ⭐ Star!