Kerog What it does Contact Get Kerog

Word by word ་ Offline ་ Free forever

Every Tibetan word, glossed, in your pocket.

  • Translator
  • Speech to text
  • Image to text
  • Dictionary
  • Text to speech

Web translators running in the cloud handle whole sentences better, and with a good signal they are worth using. Kerog is for when you do not have one: travelling on little or no data, talking with the people around you, a voice note that has just arrived.

Let it listen, share the voice note in, or photograph the page. Every sentence comes apart into its words, each with its meanings from ten Tibetan dictionaries — straight away, on the phone, with nothing uploaded. A rough English pass comes with it, and the gloss is what you check it against.

Get it on the App Store See the gloss

No account. No subscription. No analytics. Works in aeroplane mode.

The Kerog app icon: a Tibetan letter inside the five-coloured ring.
Kerog
Read ་ Hear ་ Gloss
Tibetan ↔ English
Requires
iPhone, iOS 17 or later
Bundled models
About 1 GB, all on-device
Dictionary
265,000 entries, 10 sources
Network
Never, unless you ask
Price
Free, permanently
The gloss

When the translation fails, the gloss still tells you something

A model small enough to live on a phone will get sentences wrong. So the word-by-word reading is not a footnote here — it is the point. Every sentence breaks into its words, each carrying its senses from ten Tibetan dictionaries, with the dictionary named. That is what lets you see why a translation went wrong, and work out what was really said — and because all of it is on the phone, it is there with no signal at all.

Reading Select a word

ཁྱེད་རང་གིས་སྐད་ཆ་དེ་གོ་ཐུབ་ཀྱི་འདུག

You can understand what is said.

Eight words, from the dictionaries bundled with the app. Entries as they stand, including where they differ.

Where it fits

Better translators exist. They need a signal.

Kerog does not pretend to beat the large models on a server. It is for the moments they cannot reach.

With a good connection

A translator in the cloud

Models running on servers are far larger than anything that fits on a phone, and they handle whole sentences better. At a desk with Wi-Fi, they are the better tool — and Kerog can hand them individual tasks too, through the Cloud (Monlam AI), if you switch one on in Settings.

With little or no data

Kerog already has the words

A guesthouse with no Wi-Fi, a bus between valleys, a conversation that will not wait for a page to load. The gloss, the dictionary and a rough translation are already on the phone. Listen while someone talks, look a word up by how it sounded, or type English and play it back in Tibetan.

Listening, in aeroplane mode

Nothing here is waiting on a server. Each sentence lands as it is said, takes its English and its words, and keeps its own audio so you can replay that line alone.

The name

སྐད་སྒྱུར་རོགས་པ།

skad sgyur rogs pa

The translator’s friend — not the translator.

སྐད་
ke · skad
speech, voice, language
རོགས་
rog · rogs
a friend, a helper, the one who comes along

That is the whole claim. Kerog does not translate for you, and it is not trying to. It works alongside you: the words, their senses, the dictionaries where they differ, and a rough English pass to measure your own reading against.

Which is also who it is for. This assumes you have some Tibetan, or are working at getting some — at home with a teacher, or on the road with the people you meet. It is not a way around the language but a way to stay with it — to get through this evening’s voice note or this morning’s conversation, and to see the words while you do.

The letter A within five concentric coloured rings, the image used as Kerog's icon.

The icon

This image is used in Dzogchen as well as Bön practices.

My Lama is a Kham Dzogchen master. This app exists because of his voice notes.

Why it exists

I built this to understand my Lama

He sends voice notes on WhatsApp. He sends photographs of pages. Every time, the same wall — hold the message, share it… to what? Nothing on the phone read Tibetan, and by the time I had a laptop open the moment had gone.

So it starts in the share sheet. Not a feature that was added later — the shape everything else grew around.

Hold

A voice note arrives

WhatsApp, Messages, Files, anywhere. Hold it and choose Share.

Share

Pick Kerog

The clip opens straight into the app. Photographs of a page go the same way — or use the camera.

Read

It reads it

Segmented, transcribed to Tibetan, translated, and broken into words. Nothing is uploaded.

The whole thing, in one gesture

A voice note held down in WhatsApp, shared into Kerog, and read — transcript first, then the English, then every word with its meanings. No copying, no opening the app first, no signal.

On the phone
Kerog showing a Tibetan passage broken into word chips, each with its first sense.
Word by wordEach sentence, its translation, then its words — collapsible, and never in the way.
The dictionary answering the phonetic query 'tashi delek' with Tibetan script, Wylie and senses.
Look it up by earType what you heard. Tibetan script, Wylie and phonetics all land in the same field.
The note Kerog shows on first launch: the gloss comes first, nothing leaves the phone, and the translation makes mistakes.
Said up frontThe first screen says what it is for, that nothing leaves the phone, and that the translation will get things wrong.
Five ways in

What each mode is for

Audio, Listen and Text run in both directions — Tibetan to English, or English to Tibetan.

Audio

Record, or share in a voice note from WhatsApp, Messages or Files. It is split on silence, transcribed, and translated sentence by sentence. Play it back and the sentence being spoken lights up as it goes — tap any line to jump the recording there.

Listen

Live. Transcribes and translates while someone is still talking, a sentence at a time. Each line keeps the audio it was heard in, so you can play any one sentence back on its own afterwards.

Photo

Photograph a page, or share one in from your library. Crop and rotate first if the framing was off — from the original photo, so widening a crop is always possible. Then it is read line by line, translated, and glossed. Tap the thumbnail to zoom in and compare.

Text

Type or paste Tibetan, or English. For the message that arrived as text rather than as a voice note — and for trying a sentence out while you are learning.

Look up

The dictionary on its own. Search in Tibetan script, in Wylie, in phonetics, or say the word out loud. You rarely know which of the three you are holding — the field works out which it was given.

Correcting it

Every transcript is editable, and the translation follows

A recogniser that is ninety per cent right is only useful if the last ten per cent is yours to fix. Change a syllable in any sentence, tap away, and that sentence re-translates and re-glosses itself around your correction. The whole passage above it updates with it.

Speak it back too: any line of Tibetan or English can be read aloud, at three paces. Slow stretches the syllables without dropping the pitch, so it stays the same voice — useful while you are still learning to hear where one word ends. Synthesised speech saves as an audio file you can send straight back into WhatsApp.

Watch the line above change

One syllable is fixed here. The English re-translates, the word chips re-gloss, and the whole-passage reading at the top of the screen follows — none of it asked for, all of it from tapping away.

What to expect

What to expect of it

This is the note Kerog shows on its first launch, before anything else.

The point

Every word, glossed, in your pocket

Each sentence comes apart into its words, with their meanings from ten Tibetan dictionaries. Web translators running in the cloud handle whole sentences better — but travelling on little or no data, talking with the people around you, this puts every word on your phone straight away.

Privacy

Everything happens on this phone

Listening, reading photographed pages, translating, speaking and the dictionary all run on this device. There is no account and no server: nothing you record, type or photograph leaves the phone, and all of it works in aeroplane mode.

First run

The first translation is the slow one

Translating on a phone means loading a model the size of the app itself, and that happens the first time you ask for it. Expect a wait of up to a minute or so, once. After that the model stays loaded and translations start straight away.

Accuracy

Expect the translation to get things wrong

The models are chosen to fit on a phone alongside everything else, and it shows. Names, rare spellings and long sentences are where they fail. Read the English as a rough first pass — the gloss underneath is what to check it against.

Optional upgrade

More accurate with the Cloud (Monlam AI)

When you do have a connection, cloud models are far larger than anything that fits here. In Settings, choose which tasks go to the Cloud — audio, translation, pages — leaving the rest on the phone. It uses monlam.ai’s free public service, so there is no key and no bill, but nothing promises it will be there: when it is busy or gone, the phone takes the task back. The gloss and everything you leave switched off never go anywhere. What gets sent, exactly.

Ongoing

This should keep improving

Open models for Tibetan are getting better quickly. New ones are tested against this app as they are released, and anything that beats what is bundled here replaces it.

The price

Free, and staying free

No price, no subscription, no advertising, no analytics, no accounts. This is a one-person project for a community too small for anyone to build a business on — which is exactly why it should cost nothing.

It is not a professional tool and does not pretend to be. It is here to help people learn, and to help people understand what their teachers are saying.

  • No cost
  • No subscription
  • No ads
  • No tracking
  • No account
  • Works offline
Get it on the App Store Email the developer
Credits

Built out of other people’s work

Kerog trains no models and compiles no dictionaries. Everything it knows was made by the people below and released for others to build on. Copyright in each stays with its author; this app converts and bundles them, and does not own them. The same list is in the app, under Settings.

Models on the phone

The dictionaries

  • Rangjung Yeshe Rangjung Yeshe Tibetan-English Dharma Dictionary 3.0 Erik Pema Kunsang, 2003.
  • Jim Valby James Valby’s Tibetan-English Dictionary James Valby.
  • Tsepak Rigdzin Tibetan-English Dictionary of Buddhist Terms Tsepak Rigdzin — revised and enlarged edition.
  • Dan Martin Tibetan Vocabulary Dan Martin, 2003.
  • Ives Waldo Ives Waldo’s glossary compilation Ives Waldo.
  • Compiled and kept in step by christiansteinert/tibetan-dictionary Christian Steinert, whose collection the app imports from.

Anki decks

Learner vocabulary, shared as decks by their authors. Where a deck names nobody, it is credited by its title.

  • The Heart of Tibetan Language, V1–V3
  • Small Steps Tibetan’s First 1000 Tibetan Words
  • LRZTP Tibetan — Ocean of Words

And

  • How a word sounds bophono, KVP scheme Esukhia. The phonetics tables are theirs; the state machine that reads them on the phone is a port of their code.
  • Line detection and English speech Vision · Speech · AVFoundation Apple — shipped with iOS.
  • The Cloud, when it is switched on Monlam AI Its hosted Tibetan models stand in for the ones on this phone whenever a task is switched over to them.

Every model and every dictionary above is somebody else’s work, released on its own terms and credited here by name. Kerog itself has no price, no subscription, no ads and nothing sold.