Everything you might want to know,
written plainly.

A voice‑first navigation app built for Indian roads — speak in any of 11 Indian languages (or mix them), get directions by landmarks instead of metres, and save places by just saying so.

No. Our routing engine is our own, built on OpenStreetMap plus community‑mapped lanes and landmarks. The voice and language models are also ours, trained on Indian code‑mixed speech that off‑the‑shelf models don't handle well.

Google Maps says "turn left in 200 metres onto unnamed road." Saarathi says "second left right after the Shiv mandir — turn tight hai, miss mat karna." Different unit of navigation. Different unit of language.

Yes — that's the whole point of Memory. Say "save this — yahi waala tailor," and it's done. Next time you say "tailor le chal," Saarathi takes you back. No pins, no folders, no rename dialogs.

Inside the cities we cover — almost certainly, for Indian drivers. Outside our cities, or for cross‑country highway trips, Google Maps is still excellent. We're not pretending otherwise. Saarathi is built for the lanes between the lanes, not the NH47.

Hindi, English, Tamil, Kannada, Telugu, Bengali, Marathi, Punjabi, Malayalam, Gujarati, and Odia. Hinglish and code‑mix work natively across all of them. See the languages page for live samples.

It means you can say "Saarathi, mujhe pickup point pe le chal — wahi jagah jahan kal gaya tha" and Saarathi understands it as one sentence in two languages, not as a broken English sentence with errors. We train on real code‑mixed transcripts from Indian conversations, not translated ones.

Pretty well — but we'd rather under‑promise. We train across regional accents (Marathi Hindi, Bengali Hindi, South Indian English, etc.) and our beta has covered seven states. There are still failure cases. Send us the audio when you find one and we'll fix it.

Yes, in order of waitlist demand. Tell us your language when you sign up. We add roughly one new language per quarter.

Bengaluru is in private beta. More cities will open as we expand. See the cities page for the rollout.

Public launch follows the beta — dates are not set yet. Join the waitlist and we'll write to you the moment your city opens. Early beta users keep their benefits after launch.

Android first — that's where most drivers in this segment are, and the on‑device voice stack runs efficiently on common Android chipsets. iOS is on the roadmap.

Android 9 or later, with a Snapdragon 660 or equivalent (most phones from 2019 onward). For iOS, we'll need iOS 16 and an A12 chip (iPhone XR/XS or later).

Onto your phone, and nowhere else. Speech recognition runs entirely on‑device. Your audio doesn't reach our servers unless you explicitly send a bug report with audio attached.

No. Not to advertisers, not to "anonymised data partners", not to ride aggregators. Ever. The product is built so we structurally can't — your saved places are end‑to‑end encrypted.

Yes. One tap. We don't keep a backup. We don't pre‑announce a 30‑day grace period. You hit delete and it's gone.

Anonymous crash logs and aggregated routing metrics (which lanes get used, which suggestions fail). No identifiers attached. You can switch telemetry off entirely in settings.

Free through the private beta. After launch we plan a free tier for personal use, a paid plan for rideshare and full‑time drivers, and a separate conversation for fleets. Final pricing will be announced before launch.

All 11 languages. All cities we cover. Memory. Landmark routing. On‑device voice. Everything that makes Saarathi different — yes, on the free tier.

Hourly earnings calculator, multi‑stop pickup chaining, side‑gate detection for rideshare addresses, night driver mode, and priority voice latency. Designed for someone who drives more than 4 hours a day.

₹49 per car per month for 20–200 cars, with a fleet dashboard and SLA on voice latency. Above 200 cars, talk to us — we offer volume pricing and a white‑label voice option. contact@saarathi-ai.app.

Voice yes, completely — runs on‑device. Maps cache for the cities you drive most after the first trip. So if you've used Saarathi in Bengaluru before, it'll keep guiding you through a tunnel or a 4G dead zone without missing a turn.

Median under 800 milliseconds from the end of your sentence to the start of Saarathi's reply. P95 around 1.2 seconds. We measure this on every release.

Lighter than streaming map apps, because we don't stream voice or imagery during navigation. Exact draw varies by device — we're benchmarking through the beta.

About 4 MB/hr after the first drive in a city (which caches the map). Subsequent drives are mostly local — we only fetch real‑time traffic deltas.

On the roadmap. We designed the UI for dashboard projection from day one — large hit targets, voice‑first interaction, glanceable HUD — so adapting it for CarPlay and Android Auto is mostly a packaging exercise.