Hermes

Hermes is my personal AI assistant. It is built on the open-source project Hermes Agent, runs on its own rented server, and I talk to it in Telegram. The idea behind it: one chat instead of ten programs.

The task

Running several projects means jumping between tools all day: code, editorial, the store, statistics. Every task starts with opening another program. I wanted to give work with one message – and in the end only make decisions.

How Hermes works

Hermes is the main orchestrator. It takes a task in the chat, hands it to specialised agents, checks the result and brings it to me for a decision. Anything that has not been checked does not count as done.

1. I write the task in Telegram.
2. Hermes breaks it down and passes it to the right agents.
3. The agents deliver a result with proof.
4. Hermes checks it and asks me: approve or send back?

Already in use: the Klariq editorial desk

Articles on Klariq are published through Hermes. My control panel Oko collects news from source feeds, prepares a draft in three languages and checks it: every number must be found in a source. Then the draft goes to Hermes – and only after my “yes” in the chat does the article go to the site.

Security

Hermes has its own access to Oko with precisely defined rights. Every action is logged, and access can be revoked at any time. Hermes can propose articles and have them approved, but it cannot change settings.

Status

Hermes is in development. Article approval runs every day; more tasks are being added step by step.