AI phone agent that makes and answers real calls.

Vanie AI is built on Google Gemini Live, which is audio-native speech-to-speech. The caller’s voice goes straight into the model and the reply comes back as voice — there is no separate speech-to-text, language model, and text-to-speech chain stitched together. Removing those hops keeps the conversation low-latency and real-time, so it feels like talking to a person rather than to an IVR.

Because the model listens while it speaks, the agent supports barge-in: interrupt it mid-sentence and it stops, listens, and adapts. Voice-activity detection handles natural turn-taking — knowing when the caller has finished a thought versus simply paused. Every call produces a full input and output transcript, so each turn is captured as searchable, readable text on the dashboard.

The whole agent is configurable from the dashboard: persona, prompt, model, voice, the tools it may call, transfer rules, and working hours. Changes go live within about a minute — no redeploy. You can rehearse any change in the browser with the built-in WebRTC live-test before a single real call lands.

Connect the carrier you already use: bring your own Twilio, Exotel, or any SIP trunk — with one-click presets for Jio, Vi, Airtel, Tata, and Servetel. Prefer not to manage a carrier? Buy a number from Vanie AI as a reseller. Either way, a number binds to an agent for inbound calls and acts as the caller-ID for outbound.

There is no telco migration. Keep your existing numbers and IVR, point the route at Vanie AI, and the agent picks up. Inbound and outbound run through the same agent and the same configuration, so what you test in the browser is exactly what callers hear.

Upload your documents or import them straight from Google Drive. Vanie AI ingests them in the background so large libraries do not block setup, and the agent draws on them to answer mid-call. Per-plan quotas keep storage predictable.

Mid-call, the agent retrieves the relevant passages from that material and answers from your content — pricing, policies, FAQs, product details — instead of guessing. Update a document and the agent reflects it without a redeploy, so its answers stay as current as your source material.

Schedule outbound work three ways: fill in a form, upload a CSV of contacts, or call the API from your own system. A background poller then dials each contact through the same agent that handles your inbound calls — so qualification, reminders, and follow-ups all use one consistent persona and knowledge base.

Every dial respects compliance automatically: an NCPR/DND pre-dial check and the default 10am–9pm calling window run before the call connects, with the TRAI disclosure preamble and consent capture handled in-conversation.

Turn on optional call recording and audio is stored on Cloudinary behind private, signed URLs. When a call ends, post-call processing extracts structured fields, writes a summary, and scores sentiment — then files all of it against the caller in Vanie AI’s own built-in CRM (Customer 360).

Each customer profile auto-builds from activity: interactions, full transcripts, recording links, a sentiment trend, notes, and tags. This is the platform’s own CRM — there are no external CRM connectors. To move data into your stack, Vanie AI sends HMAC-signed outbound webhooks you can verify and route wherever you like, and the dashboard surfaces usage, cost, and call metrics directly.

Pricing is usage-based at 1 credit = ₹1. On the Platform plan, Vanie AI pays the Gemini bill: a voice call is 5 credits/min, plus 0.5 for recording and an optional 2 for extraction — about 5.5/min base and 7.5/min all-in. On BYO, you bring your own Gemini key, pay only a thin platform fee, and the call drops to 1 credit/min (extraction 1.5) — saving roughly 75–95% on credits. Keys are encrypted, with a per-call key-fingerprint audit.

Illustrative example: 1,000 minutes of talk time. On Platform at 7.5/min all-in (5 call + 0.5 recording + 2 extraction) that is 7,500 credits ≈ ₹7,500; the same volume on BYO at 3/min all-in (1 call + 0.5 recording + 1.5 extraction) is 3,000 credits ≈ ₹3,000 in usage before the thin platform fee. For context, a fully-loaded 24/7 human seat typically costs many times more, while usage-based pricing means you pay only for actual talk time. Start on the Free Trial (₹0, 500 credits, no card); plans run Starter ₹999 (≈ $12) through Scale ₹9,999 (≈ $116) and Enterprise custom, plus 18% GST for India billing — pay in INR or with international cards. See the full breakdown on the pricing page.