TechnicalNews

Always On Agentic AI? 🧠 Hermes Supercomputer Setup (Abacus AI)

Wanderloots

Callum demonstrates how to deploy and run the Hermes AI agent 24/7 in the cloud using Abacus AI's Supercomputer, a managed cloud platform that eliminates the need for personal computer resources. The setup process is automated, requires just one click to install, and includes features like memory management, daily automations, and integration with messaging apps like Telegram.

Summary

The video is a tutorial on deploying the Hermes agentic AI in Abacus AI's Supercomputer cloud environment. Callum explains that agentic AI differs from chatbots because agents can take actions, not just provide answers. Running agents locally presents challenges: 24/7 device availability, model costs, security concerns, and communication infrastructure. Supercomputer solves these problems by providing a dedicated virtual machine isolated from the user's personal computer.

The installation process is remarkably simple—a single click deploys Hermes with full terminal access, web browsing capabilities, and AI agent functionality. The platform supports multiple agents (Hermes, Open Claude, Claude Code) and integrates with Abacus AI's Route LLM, which automatically selects the best model for each task. The dashboard provides access to model configuration, memory management, skills creation, and plugin integration.

Callum demonstrates practical features including setting up memory profiles, creating a daily briefing automation that researches Agentic AI news, and connecting to Telegram for direct phone notifications. The Hermes agent can create its own skills and automations through natural language requests. The memory system allows the agent to remember user preferences and improve over time through feedback loops. Users can configure auxiliary task models separately (e.g., using cheaper models for specific functions) and even implement a "mixture of agents" approach where multiple models collaborate.

The pricing structure uses a credit-based system: Supercomputer costs 1 credit per 5 minutes of operation. Running 24/7 for a month costs approximately 9,000 credits. The Pro tier ($20/month) provides 30,000 credits, meaning a fully operational agent uses less than one-third of monthly credits, leaving capacity for API calls and model usage. Users can save credits by shutting down when inactive and restarting as needed.

Key Insights

  • Agentic AI differs fundamentally from chatbots because agents can perform actions autonomously, requiring 24/7 device operation, payment infrastructure, security isolation, and communication channels—all of which cloud platforms like Supercomputer aim to solve
  • Hermes incorporates a self-evolving memory system that automatically updates user profiles through a dedicated memory tool, allowing the agent to improve across multiple sessions by retaining information about user preferences and feedback
  • Hermes can autonomously create its own skills and workflows in response to natural language requests, including writing its own automation logic and managing scheduling, which the user can then refine through feedback
  • Supercomputer supports 31 different messaging channels (Telegram, Discord, Slack, WhatsApp, Signal, iMessage, etc.), allowing users to control and receive agent updates from multiple platforms without visiting a web dashboard
  • Running an agent 24/7 on Supercomputer costs approximately 9,000 credits monthly at 1 credit per 5 minutes, leaving two-thirds of the Pro tier's 30,000 monthly credits for model API calls and other operations

Topics

Agentic AI deployment in cloudAbacus AI Supercomputer platformHermes agent configuration and setupAI agent memory and self-improvementAutomation and scheduling with cron jobsTelegram integration for messagingModel selection and optimizationCloud pricing and credit systemSecurity through isolation

Transcript

[0:00] And here we go. That's it. It just literally took one button for me to click install. Everyone is talking about Agentic AI, which makes sense. Chatbots give you answers, agents can actually do things. But running an agent yourself can be complex and expensive. You need a device that can run 24/7, a way to pay for the model that powers the agent, security systems to protect your information and keep the rest of your computer safe from an agent going rogue, and a way to communicate with the agent. For many people, that can feel overwhelming. What we need is a way to simply and safely get your agent going. While connecting it to the most powerful…

Full transcript available for MurmurCast members

Sign Up to Access

More from Wanderloots

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.