LlamaCalls: AI agents that join your calls. Try it now
Tutor

An AI tutor that asks the question back

A tutor students talk to, out loud, on a call. You decide what it teaches, how it moves through the topic, and the one thing it must never do: give the answer away.

Tutor on a live call

SSam
Rebecca

A tutor that teaches the idea, not the rule

It works on a whiteboard, asks the question back instead of handing over the answer, and only moves on once the student has found it.

Updated

Ask most AI chat tools a homework question and you get the answer, neatly formatted, in about three seconds. The student copies it down and learns nothing. Good tutors do the opposite. They ask a smaller question, wait, and let the student find the next step.

A LlamaCalls tutor works the second way, and it does it out loud. The student talks to it on a call, with a voice and, if you like, a lifelike face. You write the rules it teaches by, including the one that matters most: don’t hand over the answer.

Why a spoken tutor

Typing is a filter. A student who is stuck often can’t put what they don’t understand into a written question, so they type “explain this” and get a lecture back. Talking is different. They can think out loud, get halfway through a sentence and stop, and the tutor can hear where they got stuck.

It also keeps them in the conversation. A tutor that asks “what happens if you double the width?” and then waits gets an answer. A wall of text gets skimmed.

Teaching by asking

It’s the Socratic method: the tutor asks the next question instead of giving the answer, so the student does the thinking.

You set the tutor up by describing it in a sentence: “A maths tutor for 15-year-olds working through quadratic equations, patient, who never gives the final answer.” LlamaCalls drafts a role card from that and you edit it. For tutoring, three parts do most of the work.

Steps of the session

The order the lesson runs in. For example: find out what the student already knows, work one example together, let the student try one alone, then check they can explain it back. The tutor moves through them in that order.

Rules

Things it always does and never does. Rules that work well for tutoring:

  • Never state the final answer. Ask a smaller question instead.
  • When the student is wrong, ask them to check one step rather than saying they’re wrong.
  • Only move on once the student has explained the idea in their own words.
  • Keep each turn short. One question at a time.

Tone and mood

A tutor for a nervous 10-year-old and a tutor for a university student revising for finals should not sound the same. Set the tone, and the tutor holds it for the whole session.

What a session sounds like

Here’s the kind of exchange the rules above produce:

Student: I don’t get why the answer is negative.
Tutor: Okay. What did you get when you multiplied the two brackets out?
Student: x squared minus two x minus eight.
Tutor: Good. So which two numbers multiply to minus eight?

Nothing clever is happening there. The tutor just won’t skip ahead, which is what a patient human tutor does and what a chatbot answering a question almost never does.

Language practice

Speaking is the hardest part of learning a language to practise alone, and the part apps cover least. A LlamaCalls tutor speaks and listens in any of 13 languages: English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian, Dutch, Chinese, Malay and Tamil.

Set it up as a conversation partner at the learner’s level. “A friendly Spanish speaker from Madrid who talks slowly, uses simple vocabulary, and gently repeats a sentence back correctly when the learner makes a mistake.” The learner gets a real spoken conversation, at their pace, as often as they want.

Who builds tutors on LlamaCalls

WhoWhat they build
Tutoring businessesA tutor per subject for practice between human sessions
Language teachersA conversation partner per level, in the language being taught
Edtech teamsA spoken tutor to test with students before building their own
ParentsA patient helper for one topic their child is stuck on

Reviewing a session

Every session is saved with a full transcript, and a video recording if you turn it on. A teacher or parent can read where the student got stuck and how they got unstuck, which says more than a score does.

What it costs

A voice session is $0.03 a minute, so an hour costs $1.80. With the lifelike face it’s $0.08 a minute, or $4.80 an hour. No subscription: you top up credit, it never expires, and sessions are billed by the second.

Getting started

  1. Sign up and describe the tutor in a sentence: subject, level, and how it should teach.
  2. Edit the steps and rules. Make “never give the final answer” one of them.
  3. Choose a language, a voice, and a face or voice only.
  4. Start a session from the browser and let the student talk.
FAQ

Questions, answered

Can an AI tutor stop itself from giving away answers?

Yes. On LlamaCalls you write the tutor’s rules yourself, such as "never state the final answer; ask a smaller question instead", and it keeps to them for the whole call.

Is it good for language practice?

It is one of the better uses. The tutor speaks and listens in any of 13 languages, so a learner gets a real spoken conversation at their level rather than typing into a chat box.

Does the tutor have a face?

If you want one. Each tutor can have a lifelike video face, lip-synced in real time, or be voice only, which is cheaper.

How much does an AI tutoring session cost?

A voice session is $0.03 a minute, so an hour costs $1.80. With the video face it is $0.08 a minute, or $4.80 an hour.

More ways teams use LlamaCalls

Make your first call today

Sign up, describe your agent and ring it.

Browser and Google MeetA lifelike face or just a voiceYour tools, your accounts