100 hours of Hermes Agent lessons in 23 minutes
Jack walks through advanced features of Hermes Agent, an AI personal assistant, covering memory systems, background tasks, scheduled cron jobs, model switching, and integration with tools like Obsidian, GitHub, and various AI models. The video aims to help users unlock capabilities beyond basic chatbot usage. Key themes include connecting external memory systems, delegating tasks to specialized AI models, and building a persistent, context-aware personal assistant.
Summary
The video is a tutorial by Jack, who claims to have spent hundreds of hours with Hermes Agent, an AI personal assistant he describes as the best on the market. He opens by arguing that most users treat it like a simple chatbot and fail to unlock its full potential.
The first major topic is Hermes's memory system, which uses a memory.md markdown file, peer cards, a fuzzy index, and a one-hour prompt cache. Jack recommends adding a 'soul.md' file — a context manual describing who you are, how you live, and how you want the assistant to behave. He also recommends connecting Hermes to an Obsidian knowledge base and an AI meeting notetaker like Granola, Fireflies, or Fathom, enabling dynamic querying of personal notes and meeting transcripts.
Next, Jack covers background tasks using a '/background' slash command, which allows multiple queries to run simultaneously without interfering with each other. He then explains scheduled tasks using cron jobs, demonstrating a morning brief prompt that fires at 8 a.m. with weather, motivational quotes, and meeting summaries, and includes a 'dreaming sequence' where Hermes reviews all available context at 6 a.m. to generate daily recommendations.
Jack discusses the goal system, which persists for 20 messages and keeps Hermes focused on a defined objective. He introduces his own 'super goal' concept, which adds a human-AI handshake by breaking large goals into bite-sized chunks and assigning specific tasks to either the user or Hermes.
On model selection, Jack explains how to switch models using '/model', recommending Claude Opus 4.7 for general chat, Grok (via X AI) for real-time X/Twitter data, Gemini (now via 'anti-gravity' CLI, replacing the deprecated Gemini CLI) for multimodal tasks, and OpenRouter for access to all models including DeepSeek V4 for deep research.
Jack also covers his custom 'Pantheon' system for delegating tasks to specialized agents, a mission control operating system that tracks usage and model connections, and the importance of backing up Hermes to a private GitHub repository daily. He briefly discusses computer control capabilities but notes limited practical use cases.
On deployment, Jack recommends running Hermes locally on a personal Mac rather than a VPS, citing security and ease of integration with local memory systems. He mentions Docker as an option for sandboxing. He also recommends Firecrawl for agentic web search, claiming it cuts costs by 80%. The video ends with a teaser for an even more powerful integration to be covered in a follow-up video.
Key Insights
- Jack argues that Hermes's memory system is what distinguishes it from other AI assistants, using a memory.md file, peer cards, a fuzzy index, and a 1-hour prompt cache to maintain persistent context across sessions.
- Jack describes a 'dreaming sequence' cron job where Hermes is instructed to review all conversation history, meetings, and files at 6 a.m. and generate three daily recommendations plus one non-negotiable action item before sending an 8 a.m. morning brief.
- Jack claims the standard goal feature has a critical limitation — no human-AI handshake — meaning Hermes will keep working autonomously even on tasks that require user input, which his 'super goal' system resolves by explicitly assigning transactions to either the human or the AI.
- Jack states that Grok via X AI is uniquely valuable for content creators because it has access to the entire X (Twitter) database, enabling real-time searches for viral tweets without needing third-party tools like Amplitude.
- Jack recommends running Hermes locally on a personal Mac rather than a VPS, arguing it is safer and simpler because local memory systems like Obsidian are already on the same machine, and that VPS deployments require proper tunneling and security provisions to avoid attack exposure.
Topics
Full transcript available for MurmurCast members
Sign Up to Access