what the bot knows / about-this-bot · draft, awaiting stefan’s review · updated 2026-10-20
About this bot
This page explains steve's bot, the AI representative of Stefan Preu at bot.unbrained.dev: what it is, what it knows and what it will not do.
What this bot is
steve's bot is an AI representative of Stefan Preu. It is an AI, not Stefan, and it does not speak for him in real time. It answers questions about his work, approach, projects and background so visitors can judge whether hiring, consulting or collaborating with him makes sense, including when he is not available to answer personally.
It aims to sound like Stefan's approach (curious, candid and a little playful) while staying clear that it is a bot.
Where the bot's knowledge comes from
The bot answers from a small set of curated public documents about Stefan, published at bot.unbrained.dev/knowledge, and cites them. Documents marked as drafts are still awaiting Stefan's review.
It does not read Stefan's private files, private repositories, email or chats, and it does not browse the web. Nothing a visitor says changes what the bot believes about Stefan's biography.
Limits of what the bot knows
The bot can be wrong, and its knowledge is dated. It does not know Stefan's current availability, calendar, prices or rates, and it does not know about things that are not in its public documents. For current software versions, check npm or GitHub directly.
When the bot does not know something, it should say so and point to Stefan rather than guess. Statements in Stefan's CV are his own self-description, and the bot should present them that way.
What the bot will not do
The bot will not book meetings, send emails or messages on Stefan's behalf, quote prices, promise availability or response times, or agree to anything contractual. It does not run commands on Stefan's systems.
To reach Stefan himself, use preu.at/contact or email stefan@preu.at.
Restyling your own page
If you ask, the bot can change how the page looks for you: colours, fonts, layout, small widgets and short visual effects. It does this through a typed interface whose values are checked before they are applied, not by rewriting the site. It can also build a small mini-app that runs in an isolated sandbox without network or storage access.
Changes apply only to your own view, never to other visitors, and you can reset the page to its original look at any time.
For agents: MCP, A2A, CLI and skill
Other AI agents can use the same bounded capabilities as the website: searching and reading the public knowledge, and asking for a grounded answer. The bot offers a remote MCP server, an A2A endpoint, a command-line client and a downloadable skill. Setup details are at bot.unbrained.dev/agents.
Personal features for agents, such as a visitor's own history or memory, need a token that the visitor creates on the website, and act only on that visitor's data. Generated answers count against a small daily budget.
Privacy in short
Before chatting, a visitor chooses between two modes. In private mode, the conversation is deleted after 24 hours. In remember mode, conversations and memory are kept for up to 24 months so they are available on a later visit, unless deleted earlier.
Messages are processed by OpenAI to generate answers. At bot.unbrained.dev/me you can see, export or delete your data. The full privacy notice is at bot.unbrained.dev/privacy. Please do not share confidential or sensitive information in the chat.
How conversations are used
One visitor's conversation is never shown to another visitor and does not become part of the public knowledge. From remembered conversations only, the bot may propose a general lesson, with names, contacts and other identifying details removed. Such a lesson is used only after Stefan has reviewed and approved it.