2 min read

How I Got a Hermes AI Agent Running in Telegram via Docker Compose and a Web UI

How I Got a Hermes AI Agent Running in Telegram via Docker Compose and a Web UI

Hook

Last Tuesday, I was staring at a backlog of 47 unread support messages in my Telegram DMs. My team was drowning in repetitive questions about resetting passwords and checking order status. I needed a solution that worked without a massive rewrite of our stack.

Background

We run a small SaaS product. Our support channel is Telegram. Every day, the same three questions come in: “How do I reset my password?”, “Where is my invoice?”, “Can you extend my trial?”. My support person was spending four hours a day on these. I wanted to automate the answers without giving a third-party bot access to our internal systems.

Problem

The real pain was not the AI itself. It was the deployment. I tried a few agent frameworks before, and each one needed me to configure YAML files, edit environment variables by hand, and restart containers manually every time I wanted to change a prompt. Every change meant a new commit and a redeploy. That is not practical for a fast-moving support workflow.

Why it matters

Every hour my support person spent on password resets was an hour they could have spent on closing actual sales. The cost of doing nothing was slow response times and a growing backlog. I needed a setup that was maintainable by someone who is not a DevOps engineer.

Solution

I deployed AgentForge with a single Docker Compose file. The compose file pulled the Hermes agent image, a PostgreSQL database for state, and a web UI container. I ran docker compose up -d and had a running agent in under two minutes.

Then I opened the web UI in my browser. It gave me a form where I could write the system prompt for the agent, configure the Telegram bot token, and set the welcome message. I did not edit a single config file. I saved the changes in the UI and the agent restarted automatically.

After that, I connected the agent to my Telegram group. The agent listens to messages, interprets the intent, and replies with the correct answer. If a user asks about password reset, the agent gives the link to the self-service page. If they ask about invoices, it tells them how to find it in the dashboard.

Result

Within 30 minutes of deploying, the agent answered 12 support questions correctly. The next day, my support person handled 80% fewer password reset questions. The backlog cleared in two days.

The key insight: when you can change the agent's behavior from a web UI and redeploy without touching Docker, the bot becomes a tool the whole team can manage.

I no longer dread opening Telegram. The agent handles the repetitive stuff, and my team focuses on the real problems. That is the outcome that matters.

AgentForge

AI agents for real work. If you want to see how this works in practice, visit our product site and blog.

Share
Serhii

Serhii