AWS Builder Center

Weekend Showcase Challenge: Hermes Agent

From a simple AI assistant to an autonomous agent running on AWS

Over the summer, I explored how AI can move beyond simply answering prompts and become something that can actually take action.
My showcase project, Hermes Agent, is an autonomous AI agent designed to run tasks, interact with tools, execute workflows, and operate with minimal human intervention.
The goal was simple: instead of opening an AI chatbot every time I wanted something done, I wanted to build an agent that could work in the background and take care of recurring tasks on my behalf.

๐Ÿš€ The Vision

Most AI applications are reactive. You ask a question, the AI responds, and the interaction ends.
I wanted to experiment with something different: an AI system that can receive a goal, reason about what needs to be done, use available tools, and execute tasks automatically.
Hermes acts as a personal autonomous agent that can be extended with different tools and workflows.
For example, I experimented with scheduled workflows such as generating market briefings, monitoring prices, researching AI tools and news, and creating recurring updates.
This turns the agent from a traditional chatbot into something closer to an AI worker that can operate in the background.

๐Ÿง  How Hermes Works

At the center of the system is the Hermes agent.
A user can provide a task or configure an automated workflow. Hermes processes the request using an LLM, determines what actions are required, and then interacts with the available tools.
The workflow can be summarized as:
User / Scheduler โ†’ Hermes Agent โ†’ LLM โ†’ Tool Selection โ†’ Tool Execution โ†’ Result โ†’ Response / Notification
The agent can therefore move through multiple steps instead of simply generating a single response.
One of the biggest things I wanted to explore was the difference between an AI assistant that waits for instructions and an autonomous agent that can continue working toward a goal.

โ˜๏ธ AWS Architecture

For this showcase, Hermes is deployed on AWS.
The main AWS infrastructure provides the compute environment where the agent and its supporting services run.
The architecture consists of:
  • Amazon EC2 โ€” hosts the Hermes agent and provides the always-available compute environment.
  • AWS networking and security controls โ€” provide the environment required for the deployed agent.
  • AWS Cloud infrastructure โ€” provides the reliable environment for running scheduled and autonomous workflows.
The model layer is connected to the agent through an API-based LLM provider, while Hermes handles the agent logic, tool execution and workflow management.
The architecture was intentionally kept practical rather than unnecessarily complex. My focus was on getting an autonomous agent running reliably and then improving what it could actually do.

๐Ÿ› ๏ธ How I Built It

I started by setting up Hermes in an Ubuntu environment and getting the basic agent workflow running.
The first challenge was infrastructure.
Running an autonomous agent continuously requires more resources than simply running a small script. I had to monitor memory usage, troubleshoot processes, configure the environment, and make sure the gateway remained available.
I also experimented with different model providers and configurations before settling on a setup that allowed Hermes to communicate with the model layer reliably.
After the basic agent was working, I moved toward automation.
Instead of only asking Hermes to perform tasks manually, I experimented with scheduled workflows. This is where the project became much more interesting because the agent could perform useful work without requiring me to start every task manually.

๐Ÿ”ง Challenges

One of the biggest challenges was making an AI agent reliable enough to operate continuously.
A demo that works once is very different from a system that needs to run every day.
I had to think about:
  • Process reliability
  • Memory and compute requirements
  • API failures
  • Model availability
  • Environment variables and secrets
  • Scheduled workflows
  • Tool execution
  • Error handling
  • Monitoring the running service
Another important lesson was that giving an agent more tools does not automatically make it better.
The workflow needs clear instructions, controlled tool access and predictable execution.

๐ŸŒฑ What I Learned Across the Summer

The biggest lesson I learned during the Summer Build Series is that building with AI is not just about choosing the biggest model.
It is about designing a system around the model.
Earlier in the summer, I was primarily focused on making AI applications respond to users.
As I progressed through the different builds, I became much more interested in agents, automation, tool use and autonomous workflows.
Hermes became my opportunity to bring those ideas together.
I learned how important infrastructure becomes when an AI application needs to run continuously. I also learned that the difference between a chatbot and an agent is not simply the modelโ€”it is the combination of reasoning, tools, memory, triggers and actions around that model.
Most importantly, I learned to think about AI applications as systems rather than just prompts.

๐Ÿ”ฎ What's Next?

Hermes is still an evolving project.
My next goal is to make it more modular, add more reliable tool integrations, improve monitoring, and create a clearer interface for managing autonomous workflows.
I also want to explore more AWS-native components so that individual agent tasks can scale independently instead of relying entirely on one persistent machine.
The Summer Builds Showcase gave me a reason to stop, polish the project, and look at how far the idea had come.
What started as an experiment with autonomous AI has become a foundation for a much larger personal AI-agent system.

๐Ÿ”— Project

Sourse code - Not applicable โ€” based on an existing agent framework
๐Ÿค– Live Agent: Telegram โ€” @vpshermesddw_bot
AWS: Deployed on AWS
I'd love to hear feedback from other builders on what autonomous capability I should add next.

๐Ÿ™Œ Inspiration

I'd like to thank @diveshsutar for inspiring me during the Summer Build Series and for the ideas shared by the AWS Builder community throughout these challenges.
Building alongside other developers and seeing how everyone approached AI from different directions was one of the best parts of the summer.
#application
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article