Your personal AI employee
A personal AI agent that works for you 24/7. It doesn’t just answer — it does the work: research, files, code, price watching. Write in Telegram, get the result back.
Not another chat window. A working relationship with an agent that keeps going.
Stop repeating preferences, context and the way you like results delivered.
Not only a plan: tools, files, browser, code, a sequence of steps, a file back.
Set the job at night. Get the result in the morning, or a Telegram ping when it is finished.
Start with one boring repeating job. If it saves you 20–30 minutes a day, it’s already earning the attention it takes to set up.
At 8:30, the AI news worth reading — nothing else.
Check this price every day. Message me when it drops below €400.
Find why it crashes, fix it, run tests, say what changed.
Read 30 sources, compare the arguments, send a conclusion with receipts.
Find duplicates, pull the facts, return one file.
Open sites, compare terms, leave a shortlist.
Don’t tell me how. Make the file and send it.
Every Friday check X. Write only if something actually happened.
A typical model makes you restate the process. Hermes can keep successful ways of doing a job as its own skills. Correct it once — next time it can reuse the better process.
Say the schedule in plain language. It can run the job and deliver back to Telegram.
Every weekday at 8:00, a short plan: calendar, important mail, 3 links.
Check every hour. Write only when there is something to show.
Every Friday, look at the data and send a wrap-up.
In three days, check whether the offer moved — and say so.
Search, open pages, read files, run commands, build tables, return a document.
Find offers, open the sites, drop the bad ones, keep the rest.
Read a PDF or spreadsheet, extract what matters, send a clean report.
Inspect a repo, run tests, prepare a patch — on the agent’s machine, not yours.
The rest of the feature list, translated out of the docs.
Send a voice message instead of typing. Get the answer back as text or voice — handy when you’re not at a keyboard.
Run several profiles — one for personal stuff, one for research, one for code — each with its own memory and rules.
Ask for a market brief and Hermes can pull together a small team for it: one part takes competitors, one takes prices, one takes reviews, and it hands you back one assembled result.
Already using a service or a database? Hermes can be wired up to it directly, not limited to a fixed list of built-in buttons.
A new lead, a pull request update, a notification from a service — Hermes can wake up on an incoming event and tell you what matters, instead of waiting for you to ask.
Hermes is not a better ChatGPT — it’s a different class of thing: an agent you install, that learns your way of working, on a machine you control, with Telegram as the remote. Here’s the concrete difference:
| ChatGPT | Hermes | |
|---|---|---|
| Remembers across chats | Partly | Everything it needs to, always |
| Runs on a schedule without you | Limited (Work) | Yes, on your own server |
| Learns your process as a skill | No | Yes |
| Where you use it | A separate app | Telegram, where you already write |
| Who owns the workspace | The provider | You — it can be moved |
ChatGPT is a service you use. Hermes is an agent you set up for yourself.
No special console. Open Telegram and write the way you would write an assistant: “Check this.” “Keep going.” “What’s the status?” “Have it ready by morning.” Other channels are possible; Telegram is the one we wire first.
Instead of a temporary session, the agent gets a machine that stays up. Files, tools, skills and schedules live there. You can close the laptop. Telegram is the remote. You receive server access during onboarding — you are not locked into our UI.
Commands that delete, send or change things can wait for you.
Turn off what you don’t need. Start with briefs, monitoring and drafts.
Work can run in a contained setup rather than on your personal computer.
What it keeps about you sits in your install. Model calls still go to the provider you connect — a private server is not the same as a fully local model.
From request to your first automation — three steps, none of them technical on your side.
We’ll quote after a short demo. We usually start on a small server and scale only if the jobs need it. You pay the model provider directly.
Try a personal agent on 1–2 real jobs.
Technical floor: 2 vCPU / 2 GB RAM
Daily briefs, files, a few recurring jobs.
Technical floor: 4 vCPU / 8 GB RAM
Heavier work, code, several profiles, integrations.
Technical floor: 8 vCPU / 16 GB RAM
No. ChatGPT is excellent for quick questions and a conversation. Hermes is for jobs you want to assign: memory, files, tools, a schedule, Telegram.
Those are strong features. Short version: Hermes is your own always-on agent on a separate machine, with its own files, terminal and Telegram. See the full comparison above.
Yes — because it runs on the server we set up. ChatGPT Work can also continue some cloud tasks. The distinction is not “we uniquely work while you sleep.” It is that the workplace, skills and channel are yours.
No. You write in ordinary language. The technical surface is for the agent, so it can do more than reply with text.
We can connect ChatGPT/Codex, OpenRouter, Nous, Anthropic or another supported provider. If you have nothing yet, we suggest a simple option for your jobs and budget.
Files, memory, skills and schedules live on your install. Calls to a cloud model still go to that provider. A dedicated server is not the same as “nothing ever leaves.”
Dangerous actions can require approval. We recommend starting with briefs, analysis, monitoring and drafts.
Yes. Telegram is the B2C path: dictate or type a job, the agent runs it on the server.
That’s the intended start: a morning digest, a price watch, a file pile, a weekly report.
Yes. Hermes is open source. With a proper install you can relocate it. We hand over access during onboarding.
Tell us what you want to stop doing by hand. We’ll suggest 2–3 scenarios and show how they look in Hermes.
