Skillies.aiVoice 2

Try Voice 2.

Choose a conversation. Call or speak in your browser.

Malayalam or English. Sample details, real conversation.

Include your country code. Public demo: one free test call per day, up to 3 minutes.

OR / USE YOUR MICROPHONE

No sign-up · up to 3 minutes.

Ready when you are.

BROWSER CALL3:00
Hear a voice sample

HEAR THE VOICE

നമസ്കാരം! Skillies Voice-ലേക്ക് സ്വാഗതം. ഇന്ന് നിങ്ങളെ എങ്ങനെ സഹായിക്കാം?

Hello! Welcome to Skillies Voice. How can I help you today?

MALAYALAM VOICE AI · MADE IN KERALA

നമ്മുടെ ഭാഷ. നമ്മുടെ ശബ്ദം.

A voice that feels like home.

Malayalam-first voice AI for everyday business calls. Speak naturally in Malayalam and English.

Listen to a recorded demo, or talk to Voice 2 live in your browser. No sign-up needed.

Malayalam-first voicePublic call demosAPI access by request

ROOTED IN KERALA. HEARD EVERYWHERE.

മലയാളം MALAYALAM, AT HEARTEnglish WHEN IT COMES NATURALLY KERALA · THE GULF · EVERYWHERE

01 / കേട്ടുനോക്കൂ · TAKE A LISTEN

Everyday calls.
A familiar voice.

Hear Malayalam and English flow together in recorded, fictional examples. The recordings do not confirm bookings or send messages.

01 / Clinic5.6 s
Listen
MALAYALAM + MANGLISH

A little care.

ക്ലിനിക്കിലേക്ക് ഒരു വിളി

നാളെ രാവിലെ പത്തരയ്ക്ക് ഡോക്ടറെ കാണാനുള്ള appointment confirm ആയിട്ടുണ്ട്. ഒരു പത്ത് മിനിറ്റ് നേരത്തെ വരാമോ?

Your 10:30 appointment with the doctor tomorrow is confirmed. Could you come ten minutes early?

02 / Hotel5.6 s
Listen
MALAYALAM + ENGLISH

A warm welcome.

ഒരു യാത്രയുടെ തുടക്കം

നിങ്ങളുടെ റൂം ശനിയാഴ്ചത്തേക്ക് ready ആണ്. Check-in ഉച്ചയ്ക്ക് രണ്ട് മണി മുതലാണ്, വഴി WhatsApp-ൽ അയച്ചു തരാം.

Your room is ready for Saturday. Check-in is from 2 pm, and we'll send the way on WhatsApp.

03 / Real estate5.3 s
Listen
MALAYALAM + MANGLISH

A place of your own.

സ്വന്തമായി ഒരിടം

ആ plot കാണാൻ ഈ ഞായറാഴ്ച വരാൻ പറ്റുമോ? രാവിലെയാണോ വൈകുന്നേരമാണോ നിങ്ങൾക്ക് സൗകര്യം?

Can you come to see the plot this Sunday? Is morning or evening better for you?

04 / Delivery4.6 s
Listen
MALAYALAM + MANGLISH

On the way home.

വീട്ടിലേക്കൊരു വരവ്

നിങ്ങളുടെ order ഇന്ന് വൈകുന്നേരം എത്തും. വീട്ടിൽ ആരെങ്കിലും ഉണ്ടാകുമോ എന്ന് ഒന്ന് പറയാമോ?

Your order arrives this evening. Will someone be at home?

Play a line, then cut in while it speaks.

02 / THE CONVERSATION

A conversation.
A natural rhythm.

Two short Malayalam clinic examples. A recorded caller speaks with our Voice 2, with an edited pause between the question and the reply.

8.75s

Example length

Recorded caller. Synthetic reply. Edited example.

Caller · recorded voiceVoice 2 · syntheticStaged clinic conversation
Caller

എനിക്ക് പനിയുണ്ട്. ഇന്ന് ഡോക്ടറെ കാണാൻ സമയം കിട്ടുമോ?

I have a fever. Can I see the doctor today?

Voice 2

ഇന്ന് ജനറൽ ഫിസിഷ്യൻ ഉണ്ട്. ഏത് സമയമാണ് നിങ്ങൾക്ക് സൗകര്യം?

The general physician is available today. What time suits you?

03 / FROM HELLO TO WHAT’S NEXT

A good conversation.
A clear next step.

A clinic enquiry. A room for the weekend. This visual example shows the details a front desk can gather. Nothing is saved or sent to a team.

SAMPLE HANDOVER · MALAYALAM CALL

ഒരു appointment വേണം.

“I’d like an appointment.”

സമയം / Time
Today, 4:00 pm
Doctor
General physician
Reason
Fever
Status
Waiting for the clinic team
Hear the conversation

Sample only. No booking has been sent.

04 / VOICE AGENTS · EVERYDAY WORK

Your front desk.
In a familiar voice.

Pick a business and try a conversation in Malayalam or English. These are demo workflows with sample details. A business pilot is scoped around the bookings, records and handovers you need.

01 / Hotels & homestays

A warm welcome, every time.

Room enquiries, check-in questions and requests for your front desk.

“എനിക്ക് വെള്ളിയാഴ്ച രണ്ട് പേർക്കൊരു room വേണം.”
02 / Clinics

Keep the reception moving.

Sample doctor timings and appointment requests. Administrative demo only.

“നാളെ skin doctor ഉണ്ടോ?”
03 / Restaurants

An order, in your own words.

Menu questions, sample food orders and delivery quotes within 25 km by a checked road route.

“രണ്ട് chicken biryani pickup വേണം.”
04 / EMI reminders

A courteous reminder.

Explain a sample due date and help request a callback. No payment collection.

“എന്റെ EMI due date എപ്പോഴാണ്?”
05 / Order confirmation

Get the details right.

Read back an order, check the address and capture a change request.

“എന്റെ order-ലെ address മാറ്റണം.”
06 / Delivery updates

Keep customers informed.

Delivery windows, directions and requests for a team callback.

“Order ഇന്ന് എത്ര മണിക്ക് എത്തും?”
07 / Real estate

Turn interest into a visit.

Property enquiries, preferences and viewing requests for your sales team.

“കോഴിക്കോട് two-bedroom flat നോക്കുകയാണ്.”
08 / Admissions

A helpful first conversation.

Course enquiries, schedules and counselling callback requests.

“Weekend spoken English batch ഉണ്ടോ?”
09 / Customer support

Start with a listening ear.

Capture the issue and the next action for your support team.

“എന്റെ service request-ന് update വേണം.”
10 / Sales discovery

Find the right next step.

Understand the enquiry, qualify the need and capture a callback request.

“എന്റെ business-ന് voice agent എങ്ങനെ ഉപയോഗിക്കാം?”
11 / Cart recovery

Help an unfinished order.

Answer sample product questions and capture the reason an order was left unfinished.

“Cart-ൽ ഉള്ള product-ന്റെ delivery charge എത്രയാണ്?”
12 / Insurance renewals

Make renewal simpler.

Explain a sample renewal date and request an authorised adviser callback.

“Car insurance renew ചെയ്യാൻ എന്താണ് വേണ്ടത്?”
13 / Subscription renewals

Keep a good thing going.

Explain a sample renewal date and capture continuation or cancellation requests.

“എന്റെ subscription അടുത്ത മാസം തീരുന്നുണ്ടോ?”
14 / Failed delivery follow-up

Give delivery another chance.

Capture a corrected address and preferred time for the delivery team.

“ഇന്നലെ delivery miss ആയി; നാളെ വരാമോ?”
15 / COD to prepaid

A clear payment choice.

Explain approved sample payment options and capture the customer’s preference. No payment is collected.

“Cash on delivery online payment ആക്കാൻ പറ്റുമോ?”

A bank or enterprise rollout needs its own approved scripts, integration and review. The EMI example uses fictional account details.

05 / TRY IT YOURSELF

സംസാരിച്ചുനോക്കൂ. Make it your own.

Type a line and hear it, chat with the assistant, or talk to it live in your browser. The public demos are open without an API key. These demo conversations use Malayalam and English; production call languages are confirmed separately. Demos use fictional businesses; your requests stay in the demo.

100 free credits · about 5 minutes

0 / 300

What you type flows into the words around this box. Click any line to hear it.

SPEECH AND CONVERSATIONS

What the voice speaks.
What a call supports.

Malayalam comes first. Arabic, Hindi and English are also core languages for our work in Kerala and the GCC, alongside broader multilingual voice support. Speaking, listening and writing a reply are separate capabilities.

Speech generation

Send text and receive audio in the Skillies voice. Use it with your own speech recognition and language model. We review voice samples before API access is approved. Front Desk uses voice replies for Malayalam and the voice model’s supported languages; languages outside that range receive text replies only.

Request API access ↗

Complete calls

A call must also understand the caller and produce a useful reply. Our optional listening and reply integrations use third-party models. We confirm language support across the complete setup for your business; multilingual speech output alone does not establish call support.

Try the public demos

06 / PRICING · START WITH ONE WORKFLOW

A pilot that fits.
A plan you can understand.

Start with your busiest enquiry. We agree the hours, call allowance, integrations and handover before switching on your business number.

Kerala business pilot

Talk to us for pilot pricing

A focused workflow for a clinic, hotel, restaurant or local business. We quote after checking your expected call volume and operating hours.

  • Call languages confirmed for your workflow
  • One business workflow and team handover
  • Agreed connected-minute allowance
  • Setup and integrations quoted separately
Plan my pilot ↗

Developers & calling partners

₹3 per minute of speech.

After access is approved, recharge from ₹500 and use the voice through the API. You pay for the audio we generate, by its exact length. There is no subscription.

  • Malayalam-first, multilingual speech generation
  • Up to 300 characters per request
  • A free monthly allowance for approved access
  • Balance and usage in your account
Request API access ↗

Who pays for the phone number?

You can keep your own business number and carrier account where the carrier supports routing to the agent. You pay the carrier’s number rental and call charges directly. If you need us to arrange telephony, we show those charges separately in your quote. A browser demo has no telephone carrier charge.

Calling plans use connected minutes. The speech API is charged by the minute of generated audio. Taxes, setup, carrier charges and any custom integration are shown in your quote; concurrency and service hours are agreed for each pilot.

07 / DEVELOPER REFERENCE · VOICE 2

From a line of text.
To a familiar voice.

Base URL: https://skillies.ai/voice-api/v1. Keep your secret key on your server. The developer API is separate from the free browser demo.

  1. Request API access by emailing ehsan@skillies.ai. Once approved, keep your issued key on your server. Existing customers can sign in to manage their keys.
  2. Send a Bearer-authenticated request with voice_id: skillies-voice-2 and 1–300 characters of text.
  3. Save a successful response as WAV audio. Check usage with GET /usage; revoke unused keys in the console.
Voice 2 endpoints
MethodEndpointResponse
GET/voicesAvailable Voice 2 ID and speech-generation metadata; not a list of call languages
POST/text-to-speechWAV audio on success; a JSON error otherwise
POST/speech-to-textOptional add-on: the transcript of your audio and its language
POST/chat/completionsOptional add-on: the model’s reply, in OpenAI’s chat format
POST/converseOptional add-on: one conversation turn. Audio in; the transcript, the reply and, if asked, the reply in the voice out
GET/usageYour balance, your price per minute and this month’s use
Request example and response handling
curl --fail-with-body https://skillies.ai/voice-api/v1/text-to-speech \
  -H "Authorization: Bearer $SKILLIES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text":"നമസ്കാരം, നിങ്ങളുടെ ഓർഡർ തയ്യാറാണ്.","voice_id":"skillies-voice-2"}' \
  --dump-header response.headers \
  --write-out '%{http_code} %{content_type}\n' \
  --output response.body

Inspect the HTTP status and content type before playing or saving audio in your app. 200 with audio/wav returns audio; rename response.body to speech.wav only after confirming both. Other responses are JSON, even when curl succeeds. 401 means the key is missing, invalid or revoked. 402 means the balance cannot cover the request, or the free allowance is used up. 429 means wait before retrying. 503/504 means the service is busy or the request timed out; retry only if the response says retryable: true.

An account that has not added money uses a free monthly allowance, counted in characters, at up to 8 requests a minute. An account that has added money is charged for the exact length of the audio returned, at its price per minute, at up to 10 requests a minute. Each paid response carries X-Audio-Seconds, X-Charged-Paise and X-Balance-Paise.

The WAV is mono, 48 kHz, 16-bit. Send one sentence or short paragraph per request and play the audio in order. Amounts, dates and times written in digits are spoken in the language of the text. Never send automatic unbounded retries. A voice cloned for you is quoted separately.

Three optional add-ons use the same key and the same balance as the voice: listening (speech-to-text), replies (a language model) and whole conversation turns. Listening and replies use third-party models, each with its own language coverage; a speech-output language is not automatically a supported call language. They have no free allowance: they are for accounts that have added money. The voice works without them, so you can bring your own speech-to-text and model.

Optional add-on: listening (speech-to-text)
curl https://skillies.ai/voice-api/v1/speech-to-text \
  -H "Authorization: Bearer $SKILLIES_API_KEY" \
  -F file=@turn.wav -F language=ml-IN

Send the audio as a multipart field named file: uncompressed WAV, Ogg Opus (as in a WhatsApp voice note), or raw 16-bit mono PCM sent as audio/pcm with its sample_rate. language is optional: leave it out to have the language detected, or send a code such as ml-IN, en-IN or hi-IN. keyterms, also optional, is a JSON list of up to 50 terms to favour, such as the names of your business and products.

The response is JSON: text, language, language_probability, duration_seconds, model (skillies-stt-1) and request_id. The price is ₹1 per minute of audio, charged for the exact length of the audio you send. Each successful response carries X-Audio-Seconds, X-Charged-Paise and X-Balance-Paise.

A request takes up to 30 seconds and 4 MB of audio. An account can send up to 60 requests a minute. 400 means the audio or a field was refused, and the message says why; 413 means more than 4 MB; 415 means the audio is not WAV, Ogg or audio/pcm, or is a compressed WAV. 402 means the account has not added money, or the balance cannot cover the audio. 401 and 429 are as for the voice. 502, 503 or 504 means the service is busy or did not answer in time; retry only if the response says retryable: true.

Optional add-on: replies (OpenAI-compatible chat completions)
curl https://skillies.ai/voice-api/v1/chat/completions \
  -H "Authorization: Bearer $SKILLIES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"skillies-chat-1","messages":[{"role":"user","content":"നമസ്കാരം"}]}'

The request and the response are in OpenAI’s chat completions format, so an existing OpenAI client can point at the base URL above with the model skillies-chat-1. LiveKit Agents and Pipecat connect the same way, through their OpenAI plug-ins. stream: true (server-sent events) is supported, and tools, tool_choice and response_format are passed on to the model. Fields the model does not take, such as user and parallel_tool_calls, are ignored, not refused.

The price is ₹30 per million input tokens and ₹120 per million output tokens, rounded up to a whole paisa once per request. Tokens are counted by the model; if a response carries no count, as a stream may not, the charge is estimated from the text. A streamed reply is charged when it ends. A reply that is not streamed carries X-Charged-Paise and X-Balance-Paise.

max_tokens (or max_completion_tokens) can be 1 to 4096 and is 512 when you send none. A request carries up to 200 messages and about 60,000 characters, counting messages, tool calls and tool definitions. A message’s role can be system, user, assistant or tool, its content must be text, and n can only be 1. An account can send up to 60 requests a minute. 400 means the request was refused, and the message says why; 404 means it named another model. 402 means the account has not added money, or the balance cannot cover the request with the longest reply it allows. 401, 429 and 502 to 504 are as above.

Optional add-on: conversation turns (hear, reply and speak in one request)
curl https://skillies.ai/voice-api/v1/converse \
  -H "Authorization: Bearer $SKILLIES_API_KEY" \
  -F file=@turn.wav -F speak=true \
  --form-string 'messages=[{"role":"system","content":"You answer the phone for a clinic in Kozhikode."}]'

POST /converse takes one turn of a caller in one request. Send the audio as for /speech-to-text, the conversation so far as messages (a JSON list, with your system prompt), and speak=true to get the reply in the voice as well as in text. The response is JSON: transcript (its text and language), reply (its text), audio (with speak=true: a WAV in base64 and its length in seconds), talk_seconds, charged_paise, balance_paise and request_id. reply and audio are null when there is none.

A turn is charged once, at ₹5 per minute of talk time: the audio you send, at its whole length including any quiet in the file, plus the voice audio returned. The model’s reply is included in that price. A turn that is charged costs at least 25 paise. A clip in which no words are heard is still a turn: it is charged for its length, and reply is null. If the model or the voice fails after your audio was heard, the response still carries what was done, with reply_error or speech_error, and the turn is charged for the audio you sent. A turn that is refused, or whose audio could not be transcribed, comes back as an error and costs nothing. Your account’s own price per minute and minimum per turn are in GET /usage, under conversation.

A turn carries at most 40,000 characters of messages and a max_tokens of at most 512 (200 when you send none). A larger request gets 400 before anything is heard or charged; for a longer history, a longer reply or tools, use /chat/completions. A spoken reply can be up to about 31 seconds long; a longer one comes back as text only, with speech_error, and its audio is not charged. A turn can take up to a minute, and is charged when it finishes even if your client has stopped waiting. An account can send up to 60 turns a minute. 402 means the account has not added money, or the balance cannot cover what a turn is checked for before it starts: the audio you send plus, with speak=true, up to 47 seconds of voice (₹3.90 at ₹5 a minute). The charge itself counts only the voice audio returned. The other statuses are as for /speech-to-text.

Need streaming speech or a managed phone agent? Tell us about your use case so we can agree the right connection and capacity.

08 / BEFORE YOUR FIRST CALL

A few useful answers.

Does the demo make real bookings?

No. The demos use fictional businesses and sample details. They do not save or send orders, appointments, payments or messages. A business pilot can include a booking connection or team handover when that scope is agreed.

Is this the WhatsApp front desk too?

Voice 2 is the shared voice. WhatsApp replies and telephone conversations use different connections. For a hotel pilot we configure the business information, approved prompt and handover together.

Why can a reply take a few seconds?

The voice answers several requests at the same time. At a busy moment a request waits its turn, and the playground lets you cancel.

What happens to demo text and audio?

Typed text and microphone audio go to the services needed to generate a reply. Demo recording is disabled; use fictional details instead of real names, house addresses or contact information. Restaurant map checks send only a public locality or landmark for place lookup and route distance; names and private drop-off details are excluded from map requests. The browser stores a signed demo-usage cookie. Business retention, access and handover settings are agreed during setup. See our privacy policy and terms.

09 / OUR WORLDWIDE VISION

Born in Kerala.
At home, everywhere.

ഭാഷ കൊണ്ട് അടുത്ത്.

From a clinic in Kozhikode to a Malayali-run business in Dubai, a familiar conversation can make all the difference.

Our ambition is to build the best voice agents for Malayalis everywhere: natural in Malayalam, comfortable with English, and attentive to the details that matter to your business.

കേരളം KERALAഗൾഫ് THE GULFലോകമെങ്ങും EVERYWHERE

VOICE 2 / LET’S TALK

സംസാരിക്കാം.

Every good conversation starts
with a familiar hello.

YOUR BUSINESS / YOUR NEXT CONVERSATION

Talk about a business pilot ↗Explore the demos

Voice 2 demos · Explore the recorded clips or try a fictional business conversation live.