A complete system for
voice that actually answers

The Pilot is a one-time recharge for $600 — you get 10,000 minutes at 6¢/minute plus a 20% bonus (12,000 total), bringing the effective rate to 5¢/minute. Calls are billed in 30-second pulses. Includes a free phone number for the first 30 days, sub-account access, and up to 100 concurrent calls.

You purchase credits anywhere from $10 to $5,000 and burn them as you call. No commitments, no expiry, top up whenever your balance runs low. Best for teams running ad-hoc campaigns or testing new use cases.

Yes. Plug in your own OpenAI, Deepgram, ElevenLabs, Cartesia, Azure, or Sarvam keys and pay only for the orchestration. Useful for teams with negotiated rates or model-specific compliance requirements.

Yes — we provision Meta-managed SIP credentials through the WhatsApp Business API. Once approved, your existing Graine voice agent answers WhatsApp calls with zero prompt changes and full call analytics.

Hindi, Tamil, Telugu, Kannada, Malayalam, Marathi, Bengali, Gujarati, Punjabi, Hinglish, Indian-English, and Urdu — 12+ in production. We add new accents as customers request them, usually within a sprint.

Evals. You describe the scenarios the agent has to survive — an out-of-scope request, a customer who asks for a figure you never gave it, a do-not-call, an interruption — and a simulated customer runs each one against your real agent. Every conversation is scored for relevance, obedience, tool use and factual accuracy, and each mark it loses points at the exact turn that caused it. The same scoring then runs over your live calls, so the number you tuned before launch is the number you watch afterwards.

A single-prompt agent re-reads every instruction it has on every turn, so accuracy drops as the script grows — the greeting rules compete for attention with the closing rules. A multi-node agent scopes each turn to the step the call is actually in. Open any agent, choose Convert to multi-node, and we analyse the prompt and draft the node graph for you to review and edit. Your original agent is untouched until you decide to switch.

Yes. Build reusable components — product lists with pictures, plan comparisons, lead forms, info cards, calendar slots, quick actions — and reference them from the prompt. The agent draws the component instead of describing it, and the customer's tap comes back as a structured answer. On voice, WhatsApp, or any client that cannot draw it, the agent speaks the component's fallback line instead, so one prompt works everywhere.

Yes. Upload or record 6–45 seconds of clean, single-speaker audio and MynaTTS clones it instantly — no fine-tuning wait. The cloned voice appears in the agent's voice picker straight away. Only clone voices you own or have written permission to use.

Set a goal on the agent — "appointment booked", "payment commitment", whatever the business outcome is — and describe what counts. Every call after that is scored against it by the same post-call pass that fills your custom analytics, and Analytics shows goal rate and cost per goal instead of only minutes and spend. Calls that have not been scored are shown as unmeasured, never as failures.

Yes. New accounts get $5 in free credits — enough to run a full agent end-to-end and stress-test latency, voice quality, and call analytics before purchasing a plan.

DPDP Act compliant. India or USA data residency, your choice. PII (Aadhaar, PAN, phone, bank, health data) is redacted from transcripts before storage. On-prem deployment is available on Enterprise.