For companies
Train a new hire with something that asks them questions back.
Your induction material becomes modules. The new hire reads one, then presses a button and talks to a voice agent that has that module in front of it, answers from it and nothing else, and examines them out loud on what they just read. The result is written back as a completed check, so a manager can see who has actually understood the material rather than who has clicked to the end of it.
How it works
One agent per school, handed one lesson at a time.
The agent is not trained on your material once and then hoped for. The exact text of the lesson the trainee is reading is passed into the session as context at the moment they press the button, and the agent is instructed to answer from that and from your knowledge base only. When it does not know, it says which module covers it — or that nothing does.
Why voice, for this
Because a multiple-choice question does not find out whether they can explain it.
A four-option question tests recognition. Ask the same trainee to say the answer out loud and you find out in nine seconds whether they understood the procedure or memorised the shape of the right answer. That difference is the whole reason induction fails on the floor and passes on the test.
Spoken examination is also the only form of assessment that a hire with weak reading, English as a second language, or a job that never sits at a desk can complete honestly. The agent works at whatever pace the trainee sets, and it will happily explain the same paragraph four different ways without a supervisor having to be in the room.
What it is not: it is not a replacement for a competency sign-off, and it does not certify anybody to operate anything. It is the reading half of induction, done properly, with evidence.
Where you are
Site induction — week one
| Module check | Score | Result |
|---|---|---|
| 1 · Site rules and access | 100% | Passed |
| 2 · Personal protective equipment | 90% | Passed |
| 3 · Manual handling | 80% | Passed |
| 4 · Reporting a near miss | 100% | Passed |
| 5 · Lockout and tagout | 60% | Retake |
What a buyer supplies, and what is already here
| Piece | State | Note |
|---|---|---|
| The agent | Working now | Live on this deployment. Created in a Fusion ElevenLabs workspace, public, and answering from the lesson it is handed. Press “Ask the tutor” on any lesson. |
| Lesson context passed per session | Working now | The lesson text, course, module and title are sent as dynamic variables at session start. One agent serves the whole catalogue. |
| The private-agent path | Working now | /api/voice/session mints a short-lived signed URL from ELEVENLABS_API_KEY so the key never reaches a browser. Set the key and the agent goes private with no code change. |
| Your own agent | You supply | A buyer points BRAND.voice.agentId at an agent in their own ElevenLabs workspace, on their own account and billing. We do not resell voice minutes. |
| Your material as modules | You supply | Supplied by you: a document, a slide deck or a policy per module. We convert the first course as part of the build; after that the module editor is yours. |
| Knowledge base beyond the lesson | You supply | Optional. Upload your handbook to the agent's knowledge base so it can answer adjacent questions without leaving your material. |
| Spoken check written back as a module result | Wired per buyer | Built as the seam: the agent scores aloud today, and the write-back uses the same quiz-attempt endpoint the written checks use. Wired per buyer during the build. |
| Voice minutes | Wired per buyer | Billed by ElevenLabs to the buyer's own account. Not marked up, not resold, and not included in a seat price. |