Saathi

Your private Hindi-English voice assistant

Speak in Hinglish. Saathi handles the rest, privately.

Saathi remembers only what you allow, runs everyday tasks like bills, reminders and your daily brief, masks your personal data before any model sees it, and never sends, books or pays without your tap.

How it works

How Nemotron and Token Factory power Saathi

  1. 1

    You speak

    Hindi, English or Hinglish, straight from your browser mic or keyboard.
  2. 2

    We mask first

    Phone, email, Aadhaar, PAN, UPI, card and account numbers become tokens like <PHONE_1> before anything leaves for a model.
  3. 3

    Nemotron thinks

    Nemotron Nano on Nebius Token Factory routes your ask to a skill in milliseconds; tricky, multi-step asks escalate to Nemotron Super.
  4. 4

    You approve

    Anything that sends, books or pays waits as a card with the exact content and amount until you tap approve.

Models on Nebius Token Factory

  • Nemotron 3 Nanonvidia/NVIDIA-Nemotron-3-Nano-30B-A3B

    Fast router: intent + tool calls as JSON, every turn.

  • Nemotron 3 Supernvidia/nemotron-3-super-120b-a12b

    Reasoning: multi-step and ambiguous requests.

  • Qwen3 Embedding 8BQwen/Qwen3-Embedding-8B

    Memory recall over your encrypted memories.

Private by design

  • Personal data is masked before every model call and unmasked only for you.
  • Memories are encrypted at rest. View, edit or delete any of them.
  • No send, booking or payment happens without your explicit tap.
  • A hard daily spending cap keeps model usage in check.