तीव्र
Tivra
A personal AI audiobook studio.
Turn any PDF or EPUB into a natural, multilingual audiobook. Translate, narrate, look up meanings, and ask the book questions. All of it powered by AI that runs entirely on your device.
Designed for Apple Silicon. At home on every device Apple makes.
Any book.
Any language.
Yours forever.
Not a document reader. Not a TTS demo. Tivra is the listening side of the books you already own. Made for the way you actually read.
Any book
Drop in a PDF or EPUB. Tivra parses, cleans, and chunks it into audiobook-grade narration segments.
Any language
21 Indic languages with natural Indian narrators. Plus 190+ others through on-device translation.
Yours forever
Runs entirely on-device. No cloud. No subscription. No account. The book is yours; the listening is too.
Engineered around the frameworks Apple gives us for free.
Render once. Listen anywhere.
Built for every Apple device you own.
Generate audio once on your Mac with the highest-quality voices. Sync through your private iCloud container, the way Apple designed it. Listen anywhere. Your progress follows you.
The studio
A reader, a player,
and a producer.
Focus Mode on Mac shows the page and the sentence-by-sentence narration side by side. Watch the highlight chase the narrator. Click any line to jump. Tivra rewrites text for the ear. Code blocks get politely skipped. Lists become "first, second, finally." Headings get topic-change cues. So it sounds like a real audiobook, not a robot. Page rendering and playback ride on Apple's own PDFKit and AVFoundation, so it feels native because it is.
- Sentence-level read-along with auto-scroll
- Chapter navigation from PDF outlines & EPUB nav docs
- Streaming-first. Listening starts within seconds of import.
- Speed 0.5x to 2x with pitch preservation
- Gapless playback, lockscreen, AirPods, and AirPlay 2 controls
- Smart handling of code, equations, arrays, dates, and URLs
Confused?
Just ask.
Highlight a word for a meaning. Tap a paragraph for a simple-words explanation. Type a question and get an answer grounded in your book's own passages, with citations you can tap to jump straight to the spot. Apple Silicon's Neural Engine carries the inference. Your battery barely notices.
Every query runs through a local LLM on your device. Nothing is uploaded. Nothing is logged. The model only sees the excerpts you handed it, so it can't invent facts about your book. The worst it can say is "I don't see this in the book."
Safe for work. The only AI book reader you can use with confidential material.
Meaning, instantly.
Long-press any word. The local model returns a plain-English meaning in context, without ever pinging a dictionary server.
Explain like I'm five.
Tap a dense paragraph. The model rewrites it in simple words. Useful for jargon-heavy technical books and academic prose.
Search your book.
Ask anything. "Summarise chapter 3." "What does the author say about latency?" Answers cite the passages they came from.
Indic-first
Made for the languages
a billion readers speak.
Most spoken-Hindi audiobooks today either sound like a foreigner reading Devanagari or skip Hindi entirely. Tivra ships with two curated Indian-accent neural narrators reading 21 Indic languages with natural prosody, entirely on-device. Apple's text engine has always handled every script the world reads. The voices have finally caught up.
Anjali
Default narrator
अंजली
Warm, articulate, ideal for long-form non-fiction and literature.
Bishnu
Alternate narrator
बिष्णु
Deeper register, balanced pacing, suited to technical and academic reading.
Native Indic voices
Bring your own voice
Pick the engine.
Per book. Per language.
Tivra runs on-device by default. Free, private, always there. If you happen to have an ElevenLabs, Google Cloud TTS, or Azure Neural Speech account of your own, Settings lets you plug your key in for those voices too. Tivra never resells or proxies anyone's cloud. It simply respects the keys you already have.
Built in. Free. Private.
Your own cloud account, if you have one
On-device engines never touch the network. Cloud rendering sends chunk text to the provider's API, only when you ask it to, and only with the API key you provide. Apple's frameworks handle the secure storage of those keys in the Keychain.
And for translation
Four engines. Two hundred languages. Zero compromises.
Apple Translation
System framework, on-device, ~19 language pairs.
Premium prose engine
Audiobook-grade register for 19 major languages.
Wide-coverage engine
200+ languages, fallback for the long tail.
Indic specialist
Apple Silicon-optimised translation for Indic languages.
Read along
See the words.
Hear the page.
At its best on the iPad, the canvas Apple built for reading.
On your wrist
Leave the phone.
Take the book.
The phone renders. The watch caches. Step out for a run with just your watch and AirPods. The next few chapters travel with you over Apple's own WatchConnectivity, peer-to-peer, no servers involved. The Digital Crown adjusts volume. The sleep timer winds down. The book continues where you left off.
For everyone who reads
Built for every kind
of reader.
Reading is not one activity. Tivra was made with the people in mind for whom listening is the right way to read.
Audio-first readers
For readers with dyslexia, low vision, or anyone for whom listening is simply the better way to take in a book. Works with VoiceOver and Dynamic Type out of the box.
Language learners
Read in the original language, listen in your native one. Or the other way around. Every chunk re-narrated locally, no upload, no waiting on a server.
Technical readers
Engineering, science, and academic books that nobody narrates. Code blocks are skipped politely, equations are read correctly, your textbook becomes an audiobook.
Working professionals
Bankers, doctors, lawyers, consultants. Anyone whose books and documents cannot be uploaded to a cloud service. The on-device guarantee is the whole point.
VoiceOver, Dynamic Type, Reduce Motion. The accessibility settings you already use, respected.
What stays on your device
stays on your device.
After install, Tivra never sends your book text or your audio to any network endpoint. Every transformation in the pipeline runs locally, wrapped in Apple's CryptoKit and Keychain. The same primitives Apple trusts for their own data.
Under the hood
Capable.
Quiet about it.
A streaming Swift pipeline runs the whole show. Parsing, translating, narrating, encrypting. The heavy lifting happens on Apple Silicon, the chip that finally made local AI of this calibre possible on a consumer machine.
Neural narration
Audiobook-grade voices for 30+ languages, on-device.
Indian-accent voices
Natural prosody across 21 Indic languages.
On-device translation
Major languages, audiobook-quality register.
Long-tail languages
200+ languages covered, fallback routing.
Smart narration AI
Code, math, and arrays rewritten so they read like prose.
Apple CryptoKit
AES-GCM audio encryption, Keychain-backed.
~5s
to first playable chunk
0
network calls after install
21
Indic languages locally
On the way
Built next.
For the rest of you.
Coming soon
One voice, every language
The audiobook version of consistent character voice. Pick a single narrator and they read your book in any language you choose, prosody intact. Trained and rendered entirely on-device.
Coming soon
Smart notifications
Quiet, opt-in nudges that respect Focus and Do Not Disturb: your book is ready to listen, your translation finished, today's chapter is queued for your commute. Generated and scheduled entirely on-device.
Same models. Same privacy promise. Same offline guarantee.
Your library.
In your ear.
Tivra is in active development. Coming soon to the App Store for iPhone, iPad, Mac, and Apple Watch. Because that's where personal computing still feels personal.
Tivra is built and maintained by Vinod Sharma.