Quick Overview
In this build demonstration and architecture tour, tech creator Tina Huang showcases HermesOS, her custom multi-agent AI setup. The video provides a detailed breakdown of how she organizes autonomous agents to handle programming, life coaching, and company operations across local hardware and cloud servers.
Key Points
- 1.HermesOS is a multi-agent system configured to manage software engineering, personal health and productivity coaching, and business operations.
- 2.The Coder agent uses Discord to draft product requirement documents and automates development tasks using local models alongside Claude Code.
- 3.LifeBot functions via Telegram as a personal health and productivity coach by analysing data synced from Apple Health, custom desktop Pomodoro trackers, and Obsidian markdown logs.
- 4.TakoBot serves as an automated Chief Operating Officer and assistant across Discord, Notion, GitHub, and Slack to manage team status updates, documentation, and operational protocols.
- 5.HermesOS agents can be self-hosted locally on dedicated hardware such as a Mac Studio or deployed to run continuously on virtual private server infrastructure.
- 6.System health, scheduled cron jobs, and security checks across all active sub-agents are monitored through an automated WatchDog alert pipeline.
Summary
Tina Huang presents an architectural walkthrough and live demonstration of HermesOS, a custom multi-agent artificial intelligence environment designed to automate software engineering, personal productivity coaching, and internal company operations. She begins by demonstrating the Coder agent, which interfaces through Discord. In the demonstration, she prompts the bot to draft a specification for adding customizable character skins to an existing pixel-art desktop Pomodoro application built with Electron. The system generates a detailed product requirements document covering asset registries and timer alignments. Once approved, the build pipeline executes in the background and delivers an updated application featuring avocado, cat, dog, and tomato skins.
Huang next introduces LifeBot, a personal health and productivity coach accessible via Telegram. LifeBot aggregates data from three primary local and synced sources: completion logs from the custom Pomodoro app, task checklists from an Electron Typewriter app stored in an Obsidian vault, and daily step counts from Apple Health synced through iCloud. When asked to evaluate daily focus, LifeBot parses these markdown and telemetry records to identify peak focus hours, calculate average session durations, and recommend scheduling adjustments. LifeBot is powered by DeepSeek V4 Pro hosted in the cloud, with the ability to switch to a locally hosted Qwen 3.6 model running on a Mac Studio.
The demonstration then moves to company workflows managed by TakoBot, an agent acting as the operational assistant and Chief Operating Officer for Huang's company, Lonely Octopus. Operating inside team Discord channels, TakoBot indexes data across Notion databases, GitHub repositories, Slack channels, and Google Drive files. The agent sets cron-based reminders, summarises cross-functional team activity over the prior week, and drafts operational documentation, including release checklists with pull request and test coverage requirements.
Huang outlines the technical infrastructure supporting the system. While primary workloads run locally on a Mac Studio using models such as the Qwen 3.6 35B A3B parameter model, she demonstrates running persistent agent instances 24/7 using virtual private server infrastructure from Hostinger with Docker templates. System reliability is supervised by the WatchDog agent, which publishes daily and weekly automated status reports to a dedicated alerts channel. Huang concludes by previewing RevalBot, an upcoming research agent intended to continuously scrape web trends and evaluate software concepts.
Autonomous Software Engineering with Coder
Tina Huang demonstrates the Coder agent within HermesOS, which receives instructions via Discord to create software specifications and build features autonomously. In the demonstration, Coder creates a product requirements document and implements a multi-skin switcher feature for an Electron-based Pomodoro desktop application. The system coordinates tasks using a local Qwen 3.6 mixture-of-experts model on a Mac Studio, interfacing with Claude Code to execute the codebase modifications.
Productivity and Health Coaching via LifeBot
LifeBot acts as an analytical wellness and productivity coach accessed through Telegram. The agent reads data from local Obsidian logs, an Electron Pomodoro timer, a custom Typewriter task tracking app, and Apple Health step metrics synced via iCloud. Powered primarily by DeepSeek V4 Pro with local fallback options, LifeBot analyses productivity patterns, recommends optimal focus blocks, and advises on work breaks based on historical performance data.
Operations Management and Team Support with TakoBot
TakoBot operates as an automated assistant and Chief Operating Officer for Tina's company, Lonely Octopus. Interacting through team Discord channels, TakoBot connects to Notion, Slack, GitHub, and Google Drive to retrieve project status updates, set automated reminders, and generate operational standard operating procedures. This automation streamlines onboarding and maintains release protocols such as mandatory pull request approvals and automated testing checks.
Infrastructure, System Monitoring, and Future Agents
HermesOS relies on a WatchDog agent that continuously monitors background infrastructure, automated cron jobs, gateway connectivity, and disk usage across Discord alerts. Huang explains deployment options, comparing local Mac Studio execution against 24/7 cloud hosting on Hostinger virtual private servers using Docker templates. The video concludes with a preview of RevalBot, an in-development agent designed to conduct autonomous web research and generate content and software proposals.
The Bottom Line
The video provides a practical blueprint for structuring specialised AI sub-agents into a unified operational operating system across distinct communication interfaces. It establishes how local models and frontier cloud APIs can be paired with personal markdown notes, health telemetry, and team documentation to automate high-friction workflows. It leaves the specific architecture and autonomous research capabilities of the upcoming RevalBot agent to be revealed in future updates.
FAQ
What is HermesOS and how does the multi-agent system work?
HermesOS is a custom multi-agent system built to autonomously handle software development, personal productivity and health coaching, and business operations by coordinating specialised sub-agents across platforms like Discord and Telegram.
What data sources does LifeBot use to provide productivity advice?
LifeBot aggregates data from desktop Pomodoro logs and Typewriter app checklists stored in an Obsidian vault, combined with physical step counts synced from Apple Health via iCloud.
What role does the WatchDog agent play in the Hermes setup?
WatchDog manages system reliability by running daily and weekly checks on background cron jobs, server gateways, disk capacity, and agent health, reporting issues to a central Discord alerts channel.
Worth watching for
Developers, technical founders, and productivity enthusiasts interested in orchestrating multi-agent AI systems to automate software development, personal health tracking, and company operations.
- hermesos
- ai-agents
- automation
- productivity
- deepseek
- qwen