Section A · Day-wise & date-range context

Plan operations

Edits go live
across all calls

Set what's happening today, this week, this month. Mark stores closed, run offers, flag stock changes — Muskan picks it up on the next call.

Planning for

—

Offers

Stock alerts

Stores · status for today

Open Limited Special Closed

Section B · Your knowledge, your way

Knowledge base

— entries
— structured

FAQ doc, scripts, paragraphs, scanned PDFs — paste it or drop the file. We structure the rest.

0 chars · 0 est tokens

Saved entries are processed by our AI pipeline. Unstructured text becomes structured intent for Muskan within minutes.

Drop files here or browse

Accepts .txt, .md, .pdf, .docx, .csv, .json, .rtf — up to 10 MB each

Saved entries

Processing log

Every save · processed entry · timing

Section · Talk to Muskan end-to-end

Live test

no call yet
— turns

Type a caller utterance below. Watch Muskan stream a reply through Haiku 4.5 with the full cached system prompt. End the call to run post-call analytics — summary, bucket, tags, sentiment. Costs split three ways.

Context · caching $0.0000 cache writes + reads on system prompt

Intelligence · model $0.0000 turn-by-turn input + output

Analytics · post-call $0.0000 summary · bucket · tags · sentiment

STT — wired later

TTS — wired later

Total this call $0.0000 — turns yet

Start by typing a caller utterance below — try Hindi, Marathi, English, or anything mixed.

Caller phone no history loaded

Section · Talk to Muskan through the browser

Voice live

Mic → STT → Haiku → TTS
no telephony · localhost demo

Hold the mic and speak in any of English, Hindi, Marathi, or Hinglish. Muskan transcribes, thinks, and speaks back. Switch vendors live to A/B compare the bench winners.

STT engine TTS engine Voice / speaker Pace Call language (locked at Start)

STT $0.0000 browser mic → vendor

Model · caching $0.0000 system prompt cache

Model · tokens $0.0000 turn-by-turn LLM

TTS $0.0000 text → audio bytes

Total this call $0.0000 0 turns

Click "Start call" — mic stays open, just speak naturally.

Press and hold the mic. Speak. Release. Muskan replies in voice.

no turns yet

VAD tuning · advanced

Speech threshold (RMS) Silence threshold Min speech ms End-of-speech silence ms Audio path · use this if barge-in misses your voice Barge-in RMS threshold (your voice must exceed this) Mute agent · TEST MODE — TTS is generated but NOT played, so you can verify VAD detects your voice without echo interference

Lower thresholds = more sensitive (catches softer speech but also picks up background noise). Tune if VAD never triggers or triggers on background.

Section C · Calls Muskan couldn't resolve alone

Callback queue

— waiting
—

When Muskan doesn't have an answer or a caller asks for a human, she schedules a callback. The team picks it up here.

—In queue

—Today

—Complaints

—All-time

Search Reason

Loading…

Section B · Calls Muskan handled

Call log

— calls
— active follow-ups

Every call Muskan picked up — caller, language, duration, transcript, and analytics. Filter by phone or language to find a specific conversation.

—Today

—All-time

—Avg duration

—Resolved

Search Language

Loading…

Conversation intelligence

Analytics & insights

DEMO data
actionable view

What your callers are asking, when, and where. Patterns surface — turn each one into an action.

Period:

Action briefs —

Key metrics

When callers reach you

Calls by hour of day · peak in —

What they ask

Conversation intent breakdown

Store-wise signals

Where attention is needed — by location

Sentiment journey

How callers feel at call start vs end

Language mix

Of all inbound calls