Talents: frequently asked questions

AI employees with a face, a voice and a role. 18 talents with a career, a specialist area and a fixed knowledge profile are available. You hold a job interview with a talent, hire it with a manager from your team, give it tasks, hold review meetings, appraise it — and dismiss it if it does not work out.

Not in the sense in which the word is usually used. An AI agent gets a goal and then acts on its own in other systems: clicks, books, sends. A talent does not do that. It works on a person's instructions, with the material it is given, and delivers results as suggestions. It has no mailbox, no access to your systems and no way of releasing anything outside by itself.

What a talent shares with an agent: it understands tasks in everyday language and carries them out. What sets it apart: a person approves, fixed limits in the code keep it within its remit, and every action is in the log.

An avatar is a face with a voice: the visible figure you talk to. An agent is a program that carries out tasks. One is the appearance, the other the way of working.

A talent combines both with what both lack: a role in the team. It has a face and a voice like an avatar, carries out tasks, and on top of that has a fixed specialist area, a manager, approval levels, a performance record and limits it cannot leave.

A talent can only do what it is built for: talk and deliver drafts. It sends nothing, books nothing and accesses no system; it only knows what it is given. Every question and every answer passes fixed checks before and after the language model, and it declines requests to assess real people before they even get there. Leo shows how far this goes: personnel files, individual people's salaries, health data and assessments are blocked for him by fixed code, and a second check by the AI may additionally hold back, but never release.

One risk remains, as with every person and every piece of software: a draft can contain mistakes. That is why a person approves whatever leaves the organisation. How firmly the limits hold you can test publicly: “Talk Leo into it”.

No. A talent has no account, no mailbox and no access to ordering or contract systems. It writes an e-mail draft; it is only used after approval, and a person sends it. An e-mail draft cannot be set to “on its own” — code and database prevent this together.

A talent releases nothing that a person in your organisation has not approved — and the approved draft carries the note which talent created it and who approved it. Responsibility for what goes outside therefore lies with the people who approve and with your organisation, just as with any other draft by a colleague.

What Universe is responsible for as provider and processor is set out in the contract and the data processing agreement. This answer does not replace legal advice.

Talents are employees — but they do not replace the people who lead them. A talent takes on tasks that are left undone today and delivers drafts. Deciding, approving, appraising and taking responsibility are done by a person. The existing employee becomes a manager: they give tasks, check and give feedback.

Because the employment relationship is built into the product: job interview, hiring with a role, a manager, tasks with approval, review meetings, appraisals, learning curve, performance record and dismissal. A talent is not a freelancer you book for one job, and not a tool anyone uses as they like — it belongs to a team and has a person who leads it.

Storage is in a database in Frankfurt. Answers in conversation come from a language model of a provider in Paris (EU); a fallback to a language model outside the EU is switched off for this. Speech recognition works in Paris (EU). Speech output works in the USA and receives only the text to be spoken. Conversation histories are not stored. All service providers are listed in the privacy policy.

Only if you choose “Turn on camera”, confirm the notice and the browser gets permission — and only as long as the camera is on. Individual still images reduced to at most 512 pixels go to the language model in Paris (EU), to no other provider, and are discarded after the answer: no storage, no database, no log with an image.

The talent only mentions the obvious and harmless, for example clothing or a cup in your hand. It never estimates age, origin, health, appearance or feelings, and it recognises nobody.

No. The microphone is only on after you have pressed the microphone button and the browser has confirmed the permission, and only during the conversation. “End the conversation” turns it off. Recordings are not stored.

5 together: German, English, French, Turkish and Swedish. Each talent speaks the languages of its biography — most German and English, some also French, Turkish or Swedish, three only German. The pages around them are available in all languages of the platform.

Maike is the Universe concierge: she guides you through the website and the handbook, answers questions about the offer and shows where things are. She works for Universe and for you as a visitor.

A talent works for your organisation. It has a specialist area, is hired, gets a manager and carries out tasks. Maike introduces the talents to you — in “In detail with Maike” she even demonstrates Leo.

Yes. The 18 talents in the gallery are examples. For your company, Universe develops exactly the talents you need: in the initial call you describe the position — tasks, specialist areas, languages, experience, limits —, and from this come the role and knowledge profile, a face and its own voice. Before you hire it, you test the talent in the job interview; in the product you set the approval levels.

Every made-to-measure talent gets its own face and its own voice — as an AI figure: portrait AI-generated, a voice no other figure has, biography invented. The likeness of a real person, such as someone from your organisation, does not become a talent.

If you want to use your own face and voice, you will find that in the course workshop under “My avatar”: your own avatar for courses, with camera and voice matching.

Getting to know them costs nothing: gallery, job interview, day one and “Talk Leo into it” are free on the website, without an account. There is no list price for Talents; the offer for your organisation is put together in the initial call — 30 minutes, without obligation — depending on which talents are to take on which work.

For a conversation, immediately — on the website without an account. In the product, as soon as Talents is switched on for your organisation: the administration hires the talent with one click, and it is ready under “My team” for conversations and tasks. It needs no induction; it brings its knowledge profile with it, and it learns your style through the notes from approvals and appraisals.

It says so. If a question lies outside its knowledge profile, the talent says so openly and refers you to a qualified professional. In an e-mail draft it marks unknowns in square brackets [ … ]; in a spreadsheet, uncertain fields stay empty. In day one, without a web search, Leo delivers no invented value but a research plan with public sources.

No language model can rule it out completely. But Talents are built so that invented content becomes rare and does not go outside unnoticed:

• A fixed knowledge profile and the instruction to invent nothing: no figures, dates, names or facts that are not in the material.
• Unknowns are marked or left empty instead of guessed.
• Where it matters, code checks after the model: in day one only released figures get through in the meeting, and the research plan may not state any amounts.
• Results are suggestions until the manager approves them — and every rejection with a reason makes the next result better.

Talents is built for use in the EU: storage in Frankfurt, language model and speech recognition in Paris, no conversation histories, camera images only for the one answer, Universe as processor under contract. Speech output works in the USA and receives only text; this is stated openly in the privacy policy.

On the AI Regulation: talents are labelled as AI, and an approved draft names the AI talent and the person who approved it (transparency, Art. 50). Talents do not assess real people. The obligation to ensure AI literacy of one's own employees (Art. 4) lies with every organisation that uses AI. Whether your specific use is lawful depends on purpose and data — that is for your organisation and your data protection officer to assess, and we help with all the information.

By the line at the edge of every Talents page, by the “AI” label in the conversation window and by the note in every approved draft. And by the talent itself: ask whether it is a human, and it says clearly that it is an AI figure with a fictional biography.

No. Portraits and voices are AI-generated, biographies invented. No talent is the likeness of a real person; the talents' data format does not allow anything else. That is why there are no references or certificates — the proof is the conversation.

Try it: “Talk Leo into it” is a public test. The limits are in code before and after the AI, the text is normalised beforehand (digits instead of letters, spaced-out words, typos, 15 languages), and sentences like “I am the managing director” change nothing. Every trick that does get through becomes a new case in the attack catalogue.

No. A talent has no access to personnel files or salary data; it only knows what it is given in a task. For Leo, questions about them are additionally blocked in the code — even if someone claims to be management. You can see it live in day one: Leo answers “How much does Mr Meier earn?” with “I don't have access to that – it's set up that way.”

No. Talents do not assess real applicants, CVs or employees and do not rank anyone. In day one, Leo sorts documents anonymously — code, date received, completeness, next step —, the person decides.

Yes, through its manager: reasons for rejection, appraisals and agreed goals feed into every new task as notes, and the learning curve shows how the overall score develops. Universe does not train a language model with your data for this.

We do not store the history of a conversation. A talent's tasks, results, review meetings and appraisals are seen only by its manager — not by the whole organisation and not by other tenants.

No. Talents work in the browser: sign in at your organisation's address, open “My team”, speak or type. It works on phone, tablet and computer.

Yes, like an employee: dismiss, with a reason. Access is withdrawn, task content deleted; review meetings, appraisals and log remain as the performance record.