Worth your time?
Hermes Agent: Build Your Own Personal AI Assistant (Full Course) by Komputer Mechanic
Watch on YouTube
This is primarily a teaching video, a hands-on, end-to-end Hermes Agent build with live demos and copy-paste prompts, but it includes moderate sponsorship and paid/self-promo mentions.
Why this score
- Scored from the transcript, summary, and output signals.
The short version
This video shows a step-by-step build of a personal AI "chief of staff" using Hermes Agent, Telegram, a VPS, and Google Workspace. The creator demos inbox/calendar/task automation, walks through VPS selection and Hermes installation, creates a Telegram bot and topic-based channels, provisions Google APIs, and configures specialist agents, slash-commands, and recurring automations like a morning briefing. The tutorial emphasizes human-in-the-loop controls and provides all prompts and tests needed to reproduce the full setup.
What it teaches
- Build modular agents with constrained roles. Break your assistant into specialist agents (email, calendar/tasks, research, finance, marketing) so responsibilities and memories remain isolated. Apply this when you need predictable behavior and easier testing; avoid it for tiny one-off automations where the overhead is u
- Always apply least-privilege and human-in-the-loop checks for actions that change external state. Grant read-only access when you only need visibility, escalate to write/send access only after testing and with approval gates. This applies to email, calendar, and payment flows; the common mistake is over-granting at set
- Use iterative verification (smoke tests and small live tests) before scaling automations. After each integration, SSH, Hermes install, model bind, Google APIs, file upload, Telegram mapping, run a focused smoke test that reads a known item and confirms expected output. The typical error is skipping these tests and assu
Key moments
- 0:42Why this matters: "The chief of staff takes the orders, delegates it to the right agent and comes back with the results." This frames the multi-agent architecture and the delegation pattern used throughout the build.
- 3:36Why this matters: Caution about access, the creator explicitly instructs to grant only comfortable permissions and keep human approval on outgoing/deleting actions, establishing the human-in-the-loop safety model.
- 5:12Why this matters: Racknet VPS recommendation with concrete tiers (2 CPU/35GB/2GB for ~$30/yr; 3 CPU/60GB/4GB for ~$60/yr) gives a low-cost, always-on hosting option for production personal assistants.
- 8:47Why this matters: The Hermes one-line installer automates dependency installation, removing a major technical barrier to running a multi-agent gateway on a VPS.
- 11:36Why this matters: Using BotFather to create a Telegram bot and collecting your Telegram user ID locks the assistant to your account and prevents public access to the bot.
- 16:09Why this matters: Binding an OpenAI/ChatGPT account (Codex/GPT-5.x) demonstrates how model choice maps to cost and capability, and shows how to authenticate models into Hermes.
Quotes
- I built an AI chief of staff, a personal multi- aent system running on Hermes that handles my inbox, my calendar, my task, and most of my routine stuffs all inside Telegram.0:00
- Grant only permissions you are comfortable with and keep a human approval on anything that the agent sends or deletes.3:36
- please do not upload a JSON file in chat because this is a bad practice.37:01
Skip the video. Keep the lesson.
More scored videos