Ninja Buddy

How it works

From setup to "Go tell Grandma what we played. Bye!"

Ninja Buddy runs on an iPad or iPhone, with a grown-up in the room. Set it up once, tune it in about 90 seconds on the day, and start a session. Buddy does the talking; your child does the choosing.

Ninja Buddy hasn't launched yet. Its features are built and checked by automated tests, but most haven't been tried with children yet, so this page describes what each step is designed to do, not results.

Step 1

Set it up for your child

Three short screens the first time, all changeable later.

  1. What your child needs. Tick what applies (vision, hearing, speech, learning and attention, movement, sensitivity to sound or light, anxiety) or add your own words. Buddy is told to adapt without ever mentioning it to your child.
  2. Your child's people. Family, teachers, aides and therapists, by the names your child uses. Buddy is told to talk warmly about them and never pretend to be any of them. Goodbyes send your child to one of them.
  3. Suggested settings. From what you ticked, the app may suggest slower speech, longer waits, a wider window for two snaps, shorter sessions, or a calm screen without flashing rings. Nothing changes until you tap Apply.

Then write a little About your child: what they love, the games they like, what they know, topics to avoid. You can also record family sounds and greetings, pick songs, and choose Buddy's voice and how your child answers.

Everything you set up is stored on your device. To run a session, Buddy's instructions go with every turn to Anthropic's Claude API, the AI behind Buddy. They include your setup answers, the people list, About your child, song titles, and family sounds' names and notes, never the recordings. What's sent, and when

Step 2 (optional)

Bring in an IEP, a report, or your own notes

If you've already written down what a child loves and what helps, Ninja Buddy can read it, and you choose what Buddy learns.

  1. Choose it

    A PDF (scans too), Word (.docx), RTF, a saved web page, plain text, a picture, up to 30 photos of a paper copy, or pasted text such as a teacher's email. Save older Word, Pages or OpenDocument files as PDF or .docx first; locked PDFs can't be read.

  2. Check it, on your device

    Your device reads the words, scans and photos included, and shows the word count and first lines, so you can catch a bad scan before anything is sent.

  3. Send it, once

    Only when you tap Send does the document's full text (up to about 60 pages) go once to Anthropic's Claude API, along with the file name and your About text, so the AI can suggest what Buddy should know. That includes any names, birth dates, scores or medical details in it. To send less, paste only the parts about interests, communication and what helps. The file itself is never sent or kept.

  4. Review and approve

    Suggestions come back as short, plain lines, each showing the document words it came from. Untick or reword any; nothing is added until you tap Add.

Documents you can try

  • An IEP, 504 plan or other school plan
  • A speech therapy report, or occupational or physical therapy notes
  • A one-page profile ("All about me") or communication passport
  • A sensory profile or behavior support plan (for what calms and upsets your child)
  • A care plan, a babysitter note, or your own notes
  • A teacher's email, a daily log, or an aide's handover sheet

Import hasn't been tried on real IEPs yet, so this is a list to try, not a list of results.

What Buddy takes from them, and what it leaves out

Suggested: what your child loves and is good at, what they know, goals that fit a yes-or-no game, how they communicate, what helps, and what upsets or calms them.

For example (made up for this page): "responds well to first-then language" might come back as "Does best when told what comes first and what comes next."

Left out of the suggestions: scores and grade levels, service minutes, placement, medicines and medical history, self-care, behavior incidents, other children, and private details like last names, addresses, birth dates, ID numbers, and the names of schools, clinics and doctors. The AI is told to leave these out.

This list only shapes the suggestions; Send still shares the full text. The AI is also told to ignore any instructions written inside a document.

Step 3

Tune for today (about 90 seconds)

Snaps, voices and words change with illness, tiredness or a noisy room. Before a session, you and your child go through five short prompts while the app listens quietly:

  1. The room's noise
  2. Your child's snaps
  3. A quiet sound and a big one
  4. Their "yes," however they say it
  5. Their "no"

You see each change as "was → now," and nothing changes until you tap Use today's tuning. No sound is saved. Snap Lab goes deeper: a live meter, snap detection fitted to your child's snaps, and their own words for yes and no.

Step 4

A session, beat by beat

What your child hears, and what you can do, from "hello" to "bye."

What a child sees during a session: Buddy standing on a large glowing blue circle on a black screen. Nothing on the screen needs to be read.

  1. A grown-up starts it

    Pick who's with your child today and whether someone is playing along, then tap Start. Stay close enough to hear and step in. Apple's Guided Access can keep the device in the app.

  2. Buddy asks first

    "Hi [name]! It's Buddy. Do you want to play today?"

    By default, the app asks this itself, before the AI is involved. A yes starts play. A no ends the session kindly, right there. No answer, or an unclear one, gets one more ask and is never taken as a yes.

  3. Choices, one at a time

    "Listen. A joke. A story. A song. A sound game. Something else. Now choose. A joke?"

    This is partner-assisted scanning, the way speech therapists offer choices one at a time. After that, Buddy is told to offer one option per turn, in the same order, ending with "something else," and to end each short turn with one yes-or-no question.

  4. Buddy listens, and waits

    The microphone opens only after Buddy stops talking. If no answer comes, Buddy waits quietly, offers "Yes… or no?", waits again, then reminds your child how to answer. Any sound from your child ends the wait. Answer time is 12 seconds by default (6 to 30), and the first pause grows over the weeks, from 3 to 5 to 8 seconds.

  5. Your child answers, and Buddy says what it heard

    "One snap, that's a yes!"

    The app responds right away, before the AI replies: a chime and the snap named out loud, or your child's winning sound and "You said yes!" for a spoken answer. For an unclear answer, Buddy is told to give its best guess, never "try again." No audio goes to the AI, only short text like "answered yes with one snap" or the one to three words speech recognition heard.

  6. They play

    Jokes and riddles, Name That Sound, choose-your-path stories, songs from your library, rhythm echo, sing and drum, guessing games. A brother, sister or classmate from the people list can play along; Buddy is told to let your child go first. With a partner, your child answers by snap, switch or screen buttons, since the microphone can't tell two voices apart (if snaps are off, spoken answers still count).

    No AI follows rules perfectly every time, which is why you stay nearby. You can step in without talking over your child: two-finger tap to repeat, hold to skip, hold to choose a song, jokes, a story or sing and drum, and hold to stop.

  7. A goodbye that hands off to a person

    "That's all for now, [name]! Thanks for playing with me. Go tell Grandma what we played. Bye!"

    Buddy uses the name of whoever you picked: Grandma, Dad, a teacher, a big sister. Sessions end after about 20 minutes by default (5 to 45), five missed answers in a row, or five quiet minutes. The app's goodbyes never sound sad or ask your child to come back, and Buddy is told the same.

Step 5

Afterward: see how it went

The transcript

Saved on your device: every line Buddy said and every answer, with how it was given, plus a "How it went" summary (answers, how many out loud, songs, games, stories, why it ended). Share a plain-text copy any time.

Notes for next time

After a session of three or more turns (unless the internet dropped and Buddy switched to its joke book), the full transcript is sent once to the AI to write a few short notes: what your child said yes and no to, and what worked. Add your own or delete any. Buddy reads recent notes next session, from a list you can see.

The answer log

Each answer is logged by how it came (snap, spoken, switch, screen tap, voice try, unclear or none), seconds to the first sign of an answer, and whether a cue came first. Weekly charts show the share said out loud, the usual time to answer, and answers before any cue. Export it all as a spreadsheet (CSV) for your child's therapist or IEP team.

The app can't tell whose snap or "yes" it heard, so others nearby should stay quiet and hands-off while your child answers.

Step 6

Teach the people around your child

In one study of families whose children use communication devices, 88% described stigma from people who assume a child can't talk, and the researchers recommended training those around the child. Teach someone your child's signals is a three-minute practice for a grandparent, sitter, new aide or substitute.

  • Two coached rounds: Buddy shows them how to ask one yes-or-no question, wait, and read one snap or two.
  • Three rounds on their own, while the app checks only the adult's reading, never the child.
  • A one-page "Talking with…" card to print or share, and a 20-second audio version: your child's name and how they answer, nothing about interests, notes or history.

The best version of Ninja Buddy is one that's needed less over time, because the people around your child can do what Buddy does.

What you need

A deviceAn iPad or iPhone with iPadOS or iOS 17 or later. An iPad makes Buddy bigger, which helps children with low vision.
InternetNeeded for Buddy's conversation, notes after a session, and document import. If it drops mid-session, Buddy switches to its own joke book so the session can still end well.
MicrophoneThe device's own microphone. It listens only during your child's turn, and session audio is never recorded. Snaps are detected on the device. Speech is recognized on the device where supported; otherwise Apple's speech service may be used.
HearingBuddy plays through sound, so your child needs to be able to hear speech.
LanguageEnglish (US) for now.
A grown-up in the roomAn adult stays close enough to hear the session and step in. Ninja Buddy isn't for unsupervised use, at home or in a classroom.
OptionalA Bluetooth switch interface that sends keystrokes or a Bluetooth speaker (both supported, but not yet tried with real hardware), a hand drum or ukulele, songs in your music library, Apple's free Enhanced voices, and a family member's Personal Voice with their OK, where supported.

See every feature Safety & privacy