Glossary

The configured voice AI object you create and manage in Vomyra. The API and MCP tools use assistant as the canonical resource name. In the dashboard it may also appear as “agent.” One assistant corresponds to one Myra persona with its own language, voice, prompt, and phone number.

Microsoft Azure’s Speech-to-Text service. Recommended for all Indian-language deployments because it has the widest Hindi, Tamil, Telugu, and Hinglish support among the available transcription providers.

A pre-transcription filter that strips ambient noise � traffic, kitchen sounds, crowd chatter � before passing the caller’s voice to the STT engine. Improves transcription accuracy on calls from noisy Indian environments.

A call action Vomyra provides out of the box, such as ending a call or transferring to a human agent. No custom endpoint or code is needed.

A single phone conversation between a caller and an assistant. Calls can be inbound (a customer rings your number) or outbound (Myra dials the customer). Every call is recorded and transcribed automatically.

A tool that runs in the browser during a Web SDK voice session. Used for UI actions like highlighting a product, filling a form, or updating call state � not for sensitive backend writes.

When a caller mixes two languages in one sentence or one call � for example, Hindi and English in a single reply. Common across Indian phone calls. Myra handles it naturally; instruct it in the system prompt to “match the caller’s language style.”

A tool that calls your own HTTP endpoint during a call. Myra collects the required parameters from the caller, sends them to your API, and uses the response in the conversation. Use for order lookup, booking creation, CRM writes, and similar backend actions.

The TRAI Do Not Disturb registry. Indian regulations prohibit calling DND-registered numbers without explicit opt-in. Always exclude DND numbers from outbound campaigns.

India’s Digital Personal Data Protection Act. Governs how personal data � including call recordings, caller names, and phone numbers � may be collected, stored, and processed. Relevant for any Myra assistant that handles patient, customer, or financial data.

A first message that includes template variables such as {{name}} or {{phone}}, resolved at call time using caller data Vomyra has on file.

The exact text Myra speaks the moment a call connects. It should be short, warm, and specific to the assistant’s job � for example, “Namaste! Spice Garden mein aapka swagat hai. Main Myra � kaise madad karun?”

Transferring a live call from Myra to a human staff member. Triggered by a built-in transfer tool or by a phrase or intent in the system prompt (e.g., “if a caller is upset, offer to connect them to staff”).

A colloquial blend of Hindi and English commonly spoken in urban India. Myra handles Hinglish naturally; use Azure as the STT provider and instruct Myra in the system prompt to match the caller’s language style.

A call where the customer dials your number and Myra answers. The most common setup: attach a phone number to an assistant, and Myra handles every call on that line 24/7.

The AI model that reads the conversation transcript and decides what Myra should say or do next. Also called the “AI provider” in the dashboard. Examples: GPT-4.1 Mini (OpenAI), Llama (Meta), Vomyra’s own models.

The maximum number of tokens in a single Myra reply. Default is 256. Keep it low for phone calls � long responses sound unnatural and callers interrupt them. Increase to 512�1024 only for assistants that need to read out detailed information.

A protocol for developer environments that need structured tool access. Lets developer tools inspect Vomyra resources or expose internal business actions to an assistant in a controlled way. Not required for standard phone-call deployments.

The AI voice persona your callers hear. Each assistant you create is a Myra � a configured identity with a name, voice, language, and set of instructions. Callers speak to Myra; you configure assistants.

A call where Myra dials the customer. Used for lead follow-ups, appointment reminders, payment nudges, and post-service surveys. Subject to TRAI DND compliance rules.

A configuration field where you define what a successful call looks like � “appointment booked”, “lead qualified”, “order confirmed”. Vomyra uses this to tag calls in the analytics dashboard so you can track conversion rates.

India’s most widely used restaurant POS system. The Vomyra�Petpooja integration lets Myra take orders over the phone and push them straight to the kitchen workflow � no manual re-entry by staff.

A real telephone number attached to an assistant. Buy an Indian local number in the dashboard or import one you already own. Local numbers get answered significantly more often than toll-free or unfamiliar numbers.

Instructions written in the system prompt for what Myra should say when a tool fails � a timeout, a missing field, or an unavailable system. A good rejection plan keeps the call calm and gives the caller a useful next step.

The configurable pause between when the caller stops speaking and when Myra begins its reply. A small delay (300�500 ms) sounds natural; too little interrupts the caller, too much feels like a frozen line.

The transcription engine that converts the caller’s speech into text so the LLM can understand it. Also called the “transcriber.” Provider choice has a large impact on accuracy for Indian languages � Azure is recommended for Hindi and most Indian regional languages.

Plain-language instructions that define Myra’s persona, job, tone, limits, and tool behaviour for every call. The single biggest lever on call quality � a precise prompt on a mid-tier model usually beats a vague prompt on the best model.

A model parameter (0.0�1.0) that controls how varied vs. consistent Myra’s replies are. Lower values (0.0�0.3) produce consistent, predictable answers � best for support and FAQ bots. Higher values (0.7�1.0) produce more varied, creative replies. Default is 0.3.

An action Myra can take during a call beyond talking � check availability, save a lead, look up an order, or push data to a system. Tools come in three types: built-in, custom (HTTP), and client-side (Web SDK).

The engine that converts Myra’s text replies into the speech callers hear. Also called the “voice provider.” Vomyra AI and Azure TTS offer the best Indian-language voice quality.

The four-step loop that runs on every conversational turn: STT (hear) ? LLM (understand & decide) ? Tool (act, if needed) ? TTS (speak). Vomyra manages this pipeline; you configure each step from the dashboard.

An HTTPS endpoint you host that Vomyra calls when a call event fires � most importantly, when a call ends. Use webhooks to save transcripts, trigger follow-ups, or update your CRM automatically after every call.