Full Hermes Agent Tutorial (Desktop) 🧠 A Useful Agentic AI Workflow

Launching and Understanding Hermes Desktop 00:00

"Most AI tools are as capable on day one as they are at 100. They don't necessarily learn and evolve over time. Hermes is different."

• Hermes Desktop stands out because it is designed to improve over time by learning from its use, which makes it more effective as you interact with it.

• This self-training capability allows Hermes to adapt to your specific tasks and preferences without the need for manual re-training, making it a unique tool in the AI landscape.

Unique Features of Hermes 01:27

"The first is that it's an agent that grows with you. It gets smarter and better over time."

• Hermes possesses a persistent memory system that enables it to remember your projects and workflows, thus automatically generating skills tailored to your demands.

• The open-source nature of Hermes, developed by Nous Research, aims to enhance human rights and accessibility by providing easy access to powerful language models for everyone.

Setting Up Hermes for Local Use 02:36

"I want to show you how we can run this ourselves using a local model."

• To utilize Hermes effectively, you need to connect it to a local model such as Ollama, allowing you to manage your AI experiences without relying solely on cloud solutions.

• The process involves selecting a custom endpoint for Ollama, enabling Hermes to access locally running AI models rather than external providers.

Testing and Configuring Local Models 04:42

"What we need to do next is create a custom configuration where we create a separate variant of one of these models."

• When setting up a local model, it's crucial to ensure that it has sufficient context to operate effectively; Hermes requires a significantly larger context window compared to many models.

• Users can configure the model's context settings to enhance performance and ensure it meets the operational demands of more complex tasks.

Executing Tasks with Hermes 06:50

"All this information is 100% local; this is all just on my computer."

• After setting up Hermes and connecting it with a local model like Ollama, users can execute complex queries directly from their computer.

• This capability highlights the seamless interaction between Hermes and Ollama, allowing users to retrieve information and perform tasks without relying on external internet resources.

Connecting to OpenAI Models 07:27

"You can connect models like Codex or Claude Code directly here for more power."

• Users can experiment with different local models in Hermes, allowing them to find what works best for their needs.

• For those seeking greater capabilities, the option to link accounts from services such as OpenAI's Codex is available.

• The integration process involves opening settings, selecting providers, signing in to Codex, and entering connection codes to link the models to Hermes.

Profile Management in Hermes 08:37

"Each new profile you set up has a completely different Hermes environment."

• Hermes allows users to create separate profiles for different AI models, each with unique configurations and capabilities, enhancing versatility.

• Profiles can default to using either local models for privacy or cloud-based ones for more power. This allows users to switch models based on their specific needs or tasks.

• Creating profiles helps in maintaining isolated environments for various tasks, ensuring that specific configurations and workflows do not interfere with one another.

Skills and Tools in Hermes 10:13

"A skill is a reusable workflow that the agent can use."

• Hermes includes a built-in library of 71 skills that function as reusable workflows, enabling the agent to perform tasks more effectively.

• By using these skills, users can achieve higher consistency and quality in their interactions with the agent.

• There is a specific skill available for connecting Hermes to Obsidian, which could aid in writing notes or creating an LLM wiki efficiently.

Memory Management and Persistence 11:45

"This will build persistent memory across sessions."

• Isolated user preferences allow Hermes to develop understanding and adaptability over time, enhancing user experience.

• Conditional memory files like user.md and memory.md help Hermes retain information across sessions, enriching interaction quality by personalizing responses based on user's likes or preferences.

• The agent’s evolving capabilities enable it to improve and tailor workflows uniquely suited to the individual user.

Advanced Execution Settings 13:21

"You can protect the rest of your computer from any code or operations that Hermes might be doing."

• Users have the flexibility to execute code in a controlled environment by choosing to run the agent locally, in a Docker container, or through SSH.

• These settings help safeguard the computer’s main system from potential errors or unwanted operations during code execution.

• Users can also manage various configurations, budgets on data storage, and memory options to control how information is processed and retained.

Setting Up Memory and Gateway Connections 15:01

"You can also set up different gateway connections, allowing multiple devices to access a single memory bank."

• The video discusses building a memory stack for Hermes and mentions various potential configurations.

• Users can establish local and remote gateway connections, enabling different devices like laptops and desktops to access the same Hermes backend, thus creating a unified memory system.

• This setup allows users to utilize an agent that continues to improve its capabilities across multiple devices, ensuring consistency in memory and response.

Integrating Messaging Gateways 16:36

"We're creating a messaging gateway to message through platforms like Telegram, WhatsApp, or Discord."

• The tutorial suggests setting up a messaging gateway, with Telegram being the easiest option to start.

• To configure this, the user must create a bot via BotFather on Telegram and obtain a unique API token.

• After linking the bot with a Telegram user ID, users can communicate with their Hermes setup directly from their phone or other platforms.

Developing a Daily Briefing Bot 16:46

"I want a very curated information stream for myself that improves over time based on the feedback I give."

• The concept of developing a daily briefing bot that delivers AI updates via messaging is introduced.

• Users can program the agent to message them daily with relevant AI news, allowing for deeper dives into specific interests instead of broad information scanning.

• Feedback mechanisms are highlighted, where the bot learns from user input to refine and customize the information over time.

Creating Automation with Cron Jobs 21:40

"It's going to create a new skill called AI Daily Intel, which will become the brain of the daily report that gets better over time."

• The session involves creating a skill called "AI Daily Intel," which will autonomously run on a daily schedule.

• Cron jobs are introduced as a means for automation, allowing the user to set specific intervals for when the bot should deliver information.

• The agent can adjust its responses based on past interactions, enhancing the user's experience as feedback is incorporated into future briefings.

Self-Evolving Automation for Daily Reports 22:11

"Before you try running an automation as a cron job, it's better first to have it run in the chat so you can see what it'll look like before you try to automate it."

• Testing automations in a chat setting before scheduling them as cron jobs allows for initial adjustments, providing a more effective final output.

• Each automation can be reviewed in the chat interface, allowing users to identify and manage individual skills, such as skill authoring, enhancing the automation process.

• Users can create and modify skills, like the new daily intelligence markdown file, through tools like Obsidian, facilitating personalized updates before the next automation run.

Continuous Feedback Mechanism 22:54

"Each report I get, I can give feedback on what I like and what I don't like."

• Implementing a feedback loop where users can provide preferences helps the agent evolve and improve its outputs over time, leading to better-curated information daily.

• By scheduling cron jobs and linking them to feedback mechanisms, the agent can learn to manage and refine its skills autonomously based on user interactions.

Automation Setup and Functionality 23:34

"Everything you want to do in this system shouldn't be done manually."

• Users should set up their systems to function with minimal manual input, as agents can be programmed to handle tasks and learn from their performance.

• The ability to attach multiple skills to a cron job allows for customization and optimization of automation workflows, increasing overall efficiency.

Personalized Daily Briefings via Telegram 25:01

"The agent replied, 'Okay, cool. I saved this preference.'"

• Through interaction via Telegram, the agent can provide daily briefings that summarize significant news items, incorporating user feedback to enhance the relevance and quality of the information.

• This feature empowers users with a personalized feed that grows more tailored to their interests and needs with each interaction.

Advanced Skill Development and System Interconnectivity 26:51

"I have the skill updater cron job that runs before the next daily report."

• Users can develop a more sophisticated system where multiple cron jobs work in tandem; one for scheduling daily reports and another for updating skills based on prior feedback.

• Further integration with tools like LLM Wiki for knowledge management will enable users to better manage the information gleaned from daily reports, building a comprehensive knowledge base over time.