Service · AI office assistant
A personal AI assistant that keeps your office in shape: sorts the inbox, writes the morning briefing, publishes content — and only speaks up when it matters. Set up on your hardware, steered via chat.
This is how we work ourselves. Our assistant is called Hermine.
The idea
The workday starts with 40 e-mails, three “quick questions” and a stack of newsletters. A good assistant takes exactly that off your plate — and only speaks up when your decision is needed. If nothing is due, ours literally replies: [SILENT].
Our founder puts it like this: “Often I am only reachable for my AI assistants — and that is a good thing.”
What the assistant takes over
Three examples from our real operation — they run exactly like this, every day. And a fourth one we define together with you.
Every hour the assistant checks the Outlook inbox, classifies up to 100 mails into nine categories and tidies up: newsletters, invoices, bookkeeping — each into its folder. Important mail stays on top.
The assistant reads the industry news, opens primary sources, writes the morning briefing as a mail — and turns it into a bilingual article plus a LinkedIn post. 43 articles in eight weeks, without human intervention.
The assistant reaches out via Telegram or Slack — but only when something needs your attention. And you reach it any time in chat: ask, delegate, sharpen the rules.
You decide what your assistant takes over: filing documents, keeping to-do and callback lists, compiling reports, querying your systems. The skill system grows with you — capability by capability.
And the phone? That is where our AI phone assistants pick up — they combine nicely with the office assistant. Do call and test it: chances are an AI answers. That is exactly how you can tell it is good.
Control & privacy
Autonomy is only pleasant when the guardrails are right. That is why your assistant gets precisely defined permissions — not a master key.
The assistant runs on your own machine or server; memory, history and logs stay local. Reachable only via your private network.
Every capability gets only the permissions it needs — the mail triage, for instance, may file mails but on principle never reply, delete or send.
The orchestration is open source and model-agnostic: you choose the AI provider — US cloud, EU region or local — and can switch at any time.
Every run is logged: what was checked, what was filed, what was reported — including reasoning and confidence per classification.
For classification the assistant uses the AI model of your choice — data-frugal, primarily with metadata instead of mail bodies.
Included in the setup
We install, configure and hand over — these building blocks are included as standard.
hourly, nine categories, folder automation
workdays at 7:00, researched with sources
blog and social, bilingual — if you wish
Telegram, Slack or terminal, your choice
dictate instead of typing, transcribed locally
recurring tasks with a full log
learns your rules, extensible step by step
open-source agent, MCP interfaces, no lock-in
The honest benchmark: about eight out of ten assistant sessions at our place run fully automatically on a schedule — not in chat. The assistant works while we do not.
Technology
The foundation is a proven open-source agent (Hermes by Nous Research, MIT licence) that we set up on your hardware and extend with your capabilities: custom skills, custom schedules, narrowly scoped tools for your systems.
You choose the AI model behind it — the wiring is model-agnostic and switching is possible at any time. Systems are connected via open MCP interfaces: Outlook/Microsoft 365 today, your line-of-business systems tomorrow.
Deliberately no — at least not unasked. The triage sorts and alerts; replies stay with you. On request the assistant prepares drafts — sending happens only after your approval.
The assistant itself runs on your hardware; memory, history and logs stay local. For AI classification, requests go to the model provider of your choice — whether US cloud, EU region or a local model is up to you.
That is what the guardrails are for: the triage only files, never deletes, and leaves important mail in the inbox. Every classification comes with reasoning and confidence, every run is logged — and you sharpen the rules simply in chat.
The office assistant itself does not — that is what our AI phone assistants are for, and the two can be combined on request. Do call and test it: chances are an AI picks up.
With the use case that annoys you most every day — usually the inbox. We set up the assistant on your hardware, define the rules together, and then extend it capability by capability.
The open-source agent underneath is called Hermes — like the messenger of the gods. Our instance earned the name Hermine as a “Hermes agent with a licence for chaos”. Yours can be called whatever you like.
Self-experiment & contact
Everything on this page runs at our own company — daily, with a real inbox and real deadlines. Tell us what eats the most time in your office, and we will tell you honestly how much of it an assistant can take over.
AI office assistance is a service of Pabeschitz Computer Consulting GmbH based on the open-source agent Hermes (Nous Research, MIT licence). Feature scope and data flows depend on your configuration and the chosen model provider.
