Let Grok Bot, ChatGPT, Claude, or your booking software pick up the phone, call any Australian mobile or landline from your own Caller ID, speak a message in a natural human-like voice, and return the other person's exact spoken answer as text.
destination + initial_text)grok-voice-3.0 & grok-voice-transcribe-2.0SIP REFER) to Your Staffhttps://api.crazytel.io
{
"destination": "0491570156",
"initial_text": "Hi Sarah, I'm calling from Acme Solar to confirm if someone will be home at 10am tomorrow?"
}
{
"success": true,
"status": "completed",
"call_connected": true,
"callee_transcript": "Yes, I will be home at 10am.",
"closing_text_spoken": "Wonderful, thank you for confirming! We've locked in 10am tomorrow. Have a great day, goodbye!",
"analysis": {
"amd_verdict": "conversation",
"callee_reply_intent": "affirmative",
"callee_reply_confidence": 0.9912
},
"billing": {
"currency": "AUD",
"tts_character_count": 180,
"stt_duration_seconds": 3.7,
"ai_retail_cost_aud": "0.005388",
"ai_cdr_status": "billed"
}
}
CrazyAI Voice API bridges the gap between text-based AI/automation and the real telephone network - without requiring you to build SIP trunks, WebRTC media servers, or audio streaming pipelines.
Powered natively by grok-voice-3.0 (speech-to-speech) and grok-voice-transcribe-2.0 (speech-to-text), CrazyAI Voice API lets Grok Bot place real outbound phone calls in Australia and worldwide. Save your standing persona in the portal and pass call-specific goals in a single JSON request.
Give ChatGPT Custom Actions, OpenAI Assistants, or Claude (via Model Context Protocol / MCP tools) the ability to make phone calls. Your AI doesn't need to process raw audio - it simply sends a phone number and what to say, and receives back clean JSON containing the person's exact words (callee_transcript) and whether they said yes or no (callee_reply_intent).
Automate appointment confirmations, reminders, and schedule checks for dental clinics, medical practices, salons, restaurants, and professional services. Ask "Will you be attending your 10am appointment tomorrow?", automatically speak a tailored goodbye based on whether they confirm or decline, or transfer the live call straight to your reception desk if they need to reschedule.
Connect n8n, Make, Zapier, HubSpot, ServiceM8, or your custom CRM to verify site readiness before dispatching a crew - confirming someone is home, asking for gate codes, checking if there is a dog in the yard, or notifying customers that their driver is 15 minutes away.
Configure your voice profile once in MyAccount and trigger outbound AI phone calls from any application or agent with a single HTTPS request.
Choose which of your Crazytel PAYG Hybrid SIP Trunks to dial out from (or create a new Hybrid Trunk inline in seconds using your owned Australian Caller ID) and pick one of 28 natural voices. Click Play Sample in the portal to hear it immediately.
profile_uuid)When you save your Voice Profile, Crazytel gives you one unique Profile ID (profile_uuid). Your phone line, Caller ID, AI voice, voicemail rules, and polite goodbye settings are all stored inside that ID.
Whenever your AI bot or booking system wants to call someone, it sends just the phone number (destination) and what to say (initial_text) to your Profile ID. Crazytel dials the number, speaks your message when they answer, listens to their reply, and returns their exact words as text (callee_transcript).
Every Voice Profile supports three distinct call modes (default_mode on the profile, or mode per call in the API). You can set your preferred default in the MyAccount portal and switch modes on any individual call.
| Call Mode | API Value | Who Drives the Call? | How AI Usage Is Billed | Best For |
|---|---|---|---|---|
| 1. Quick Message & Reply (Default) | "single_turn" |
Crazytel Automated Turn (1 question → 1 transcribed answer → smart goodbye → hang up) |
TTS Speaking ($0.0285 / 1k chars) + STT Listening ($0.0063 / min) + optional AI AMD / Intent ($0.000181 / 1k tokens) |
Appointment confirmations ("Will you be home at 10am?"), delivery alerts, quick yes/no checks. |
| 2. Back-and-Forth Conversation | "interactive" |
Your Own AI Bot Decides (Grok Bot, ChatGPT, Claude, or your app controls every turn via /say, /dtmf, /refer, /hangup) |
TTS Speaking ($0.0285 / 1k chars) + STT Listening ($0.0063 / min) per turn |
Multi-step booking changes, IVR keypad navigation, or calls that need live transfer (/refer) to your staff. |
| 3. Full AI Phone Agent | "grok_voice_agent" |
Autonomous Grok Voice 3.0 (Talks back and forth automatically from start to finish using your instructions) |
Full AI Phone Agent S2S ($0.1520 / min; $9.12 / hr) |
Hands-off conversations where you want the AI to conduct the entire phone call by itself. |
Select a mode below to jump to its execution flow, or compare all three side-by-side.
"single_turn")The fastest, lowest-cost way to ask a customer a question and get their exact spoken answer back as text.
destination over your PAYG Hybrid SIP Trunk (optionally running Crazyswitch AI AMD to check for voicemail).
initial_text in your chosen AI voice (e.g. "Hi Sarah, I'm calling from Acme Dental to confirm your 10am appointment tomorrow. Will you be able to make it?").
callee_transcript alongside analysis.callee_reply_intent ("affirmative", "negative", or "neutral").
closing_message_mode):
conditional (Smart Goodbye Based on Their Answer): Speaks closing_text_affirmative if they said yes, closing_text_negative if they said no, or closing_text if neutral.static (One Standard Polite Goodbye): Speaks default_closing_text / closing_text before hanging up.wait_for_bot (My Own AI Bot Will Send the Goodbye): Immediately returns callee_transcript to your bot while holding the call open for closing_wait_timeout_seconds (3-60s, default 10s) so your bot can send a custom goodbye to POST .../hangup.none (Hang Up Immediately): Disconnects right after transcribing their answer."interactive")Designed for Grok Bot, ChatGPT, Claude, and custom booking workflows that want full programmatic control over a live phone call - including live transfers to human staff.
destination, speaks initial_text, and transcribes the person's first answer into callee_transcript.
awaiting_next_turn): Instead of hanging up, Crazytel keeps the call live on the line (status: "awaiting_next_turn", session.keep_alive: true) and returns ready-to-call URLs in session.actions.
callee_transcript and calls any of four live session endpoints as many times as needed:
POST .../say - Speak the next reply (text), optionally wait for a voicemail beep (wait_for_beep: true), and either listen for the next answer (listen_for_reply: true) or end the call after speaking (end_call_after: true).POST .../refer - Transfer the live call via Hybrid SIP REFER to an internal staff extension (101), call park slot (700), Australian phone number (0290001111), or sip: URI, with an optional announcement_text spoken before transferring.POST .../dtmf - Press keypad digits (0-9, *, #, A-D, ,) to navigate external IVR menus and transcribe the next prompt.POST .../hangup - Optionally speak a final closing_text and cleanly disconnect (SIP BYE)."grok_voice_agent")Let Grok Voice 3.0 (grok-voice-3.0) run the entire multi-turn phone conversation autonomously in real time - with zero per-turn API round-trips required during the call.
default_instructions): Save your standing business persona in the portal (e.g. "You are Eve, the friendly receptionist for Acme Solar. Always be polite, concise, and helpful.").append_instructions): On each API call, pass the specific task (e.g. "You are calling John Smith to confirm his 2pm roof inspection tomorrow at 42 Wallaby Way and ask if there is a gate code or dog in the yard."). Crazytel automatically combines "{default_instructions}\n\n{append_instructions}" (returned in instructions_used).instructions): Or pass instructions on a call to completely replace default_instructions for that call.initial_text: In grok_voice_agent mode, initial_text is optional whenever instructions are present.turns[]), callee_transcript, and AUD billing breakdown.
From answer-dependent goodbyes and Crazyswitch AI Voicemail Detection to native SIP REFER transfers and Australian suburb vocabulary.
closing_message_mode)Never hang up abruptly after someone answers your question. In single_turn mode, CrazyAI Voice API analyses the person's spoken reply and wraps up the call naturally:
conditional - Recommended): Automatically speaks your positive closing if they confirm ("Wonderful, we've locked in 10am tomorrow. Have a great day, goodbye!"), your negative closing if they decline ("No problem at all, we'll text you alternative times. Goodbye!"), or a neutral closing if their reply is open-ended.static): Speaks a single polite sign-off before hanging up.wait_for_bot): Immediately returns the person's transcript to your bot while holding the phone line open (3-60 seconds) so your AI can generate a custom closing message and send it to /hangup.none): Ends the call as soon as the reply is transcribed.Know whether a real human or a voicemail box answered before wasting speech synthesis - or leave a crisp voicemail after the beep:
conversation, voicemail, no_message, ivr, screener) and automatic voicemail beep detection (440-1500 Hz).leave_voicemail): Waits for the voicemail greeting and beep tone (or post-greeting silence) to finish, speaks your default_voicemail_text or per-call voicemail_text, and hangs up.terminate): Hangs up as soon as voicemail is detected ($0.00 TTS cost).report_only): Holds opening speech and returns voicemail_detected so your interactive bot can decide whether to leave a tailored message via /say with "wait_for_beep": true.nothing): Continues the call normally.POST .../refer)When a customer asks to speak to a person or needs complex rescheduling, your bot can transfer the live phone call via native Hybrid SIP REFER to:
101, 102)700)0290001111) or sip: URIsannouncement_text: "One moment please while I transfer you to our scheduling team") and optional Approved Transfer Numbers allowlist (allowed_transfer_targets) for strict security.POST .../dtmf)Need your AI bot to navigate an external phone menu ("Press 1 for Bookings, Press 2 for Accounts")? Send DTMF keypad digits (0-9, *, #, A-D, and , pauses) on any active interactive call and automatically transcribe the next voice prompt.
turns[]), and itemized AUD cost.POST notifications at your webhook_url as soon as a call completes, signed with X-Crazytel-Signature-256.en) for Australian calls, auto language detection, Chinese (zh), Japanese (ja), Spanish (es-MX), and 17 more.[pause], [long-pause], [laugh], [chuckle], [sigh], [breath], or wrap phrases in <soft>...</soft>, <whisper>...</whisper>, <slow>...</slow>, <fast>...</fast>, and <emphasis>...</emphasis>.$45.50 as "forty-five dollars and fifty cents" and 10am as "ten a-m" automatically.keyterms & pronunciation_replace)keyterms / extra_keyterms): Teach the speech-to-text engine up to 100 custom words - Australian suburb names (e.g. Woolloongabba, Indooroopilly), staff names, or product SKUs - so transcripts are spelled accurately.pronunciation_replace): Tell the AI voice how tricky brand names or suburbs sound phonetically (e.g. Crazytel = Crazy-tel) so it pronounces them clearly every time.Configure your standing defaults once in the MyAccount portal and override or extend any parameter on a per-call basis via the API (full schema in the CrazyAI Voice API Spec):
| Feature | Saved on Profile (Portal Default) | Sent Per-Call (POST .../calls) |
How They Combine at Runtime |
|---|---|---|---|
| Call Mode | default_mode (single_turn, interactive, grok_voice_agent) |
mode |
Uses default_mode unless mode is passed on the call request. |
| Full AI Phone Agent Instructions | default_instructions (or instructions) |
instructions (override) or append_instructions (append) |
instructions replaces profile instructions for that call; append_instructions combines "{default_instructions}\n\n{append_instructions}" (returned in instructions_used). In grok_voice_agent mode, initial_text is optional when instructions are present. |
Polite Goodbye (single_turn) |
closing_message_mode, default_closing_text, default_closing_text_affirmative, default_closing_text_negative, closing_wait_timeout_seconds |
closing_message_mode, closing_text, closing_text_affirmative, closing_text_negative, wait_for_closing_text, or POST .../hangup |
Per-call fields override profile defaults. Passing closing_text_affirmative/negative auto-enables "conditional"; wait_for_closing_text: true enables "wait_for_bot". Spoken text is returned in closing_text_spoken. |
| Voicemail Detection (AMD) | voicemail_detection_enabled, voicemail_action, default_voicemail_text, voicemail_confidence_threshold, voicemail_max_wait_beep_seconds, voicemail_post_greeting_silence_ms |
voicemail_detection_enabled, voicemail_action, voicemail_text, or POST .../say {"wait_for_beep": true} |
Per-call fields override profile defaults. On "leave_voicemail", speaks voicemail_text → default_voicemail_text → initial_text after the beep (returned in voicemail_text_spoken). |
| Custom Vocabulary & Pronunciation | keyterms, pronunciation_replace |
extra_keyterms, pronunciation_replace |
Per-call extra_keyterms merge with profile keyterms (up to 100 unique terms). Per-call pronunciation_replace entries merge and win on conflict. |
Built from the ground up for Australian businesses, software teams, and automation builders who want reliable voice outreach without telephony overhead.
Your developers or no-code workflows never touch RTP audio packets, WebSockets, or SIP trunks. If your app can send a POST request with two JSON fields, it can make phone calls.
If the person doesn't pick up, the line is busy, or the call is cancelled before answer, you pay $0.00 AUD in CrazyAI Voice API usage.
Calls dial out through your Crazytel Hybrid SIP Trunk using your verified Australian landline (02, 03, 07, 08) or mobile (04) number - so customers recognize your business and can call you back.
Replace hours of repetitive front-desk reminder calls with instant, polite AI check-ins that write the customer's exact response straight back into your booking system.
Configurable max call duration (default 5 minutes), max ring time, idle session timeouts, 5-call account concurrency limit, and a 5-second duplicate-destination cooldown prevent runaway bot loops from ever running up a bill.
No USD conversion guesswork - every character spoken, second listened, and AI agent minute is rated transparently in Australian Dollars (AUD).
CrazyAI Voice API has no monthly subscription fee. Connected calls have two transparent charges:
| Component | AI Model / Engine | Retail Rate (AUD) | Everyday Equivalent |
|---|---|---|---|
| Speaking (Voice Playback / TTS) | Grok Voice TTS |
$0.0285 AUD / 1,000 chars | ~Half a cent per sentence ($28.50 AUD / 1M chars) |
| Listening (Streaming Speech-to-Text) | grok-voice-transcribe-2.0 |
$0.0063 AUD / minute | $0.38 AUD / hour while listening |
| Listening (REST Speech-to-Text) | grok-voice-transcribe-2.0 |
$0.0032 AUD / minute | $0.19 AUD / hour |
| Full AI Phone Agent Mode (S2S) | grok-voice-3.0 |
$0.1520 AUD / minute | $9.12 AUD / hour ($0.0076 AUD / text input) |
| Voicemail Tone & Regex Detection | Crazyswitch Tone & Regex |
$0.0000 AUD | Free |
| Crazyswitch AI AMD & Reply Check | Crazyswitch AI AMD |
$0.000181 AUD / 1,000 tokens | $0.1809 AUD / 1M tokens |
| Unanswered / Busy Calls | - | $0.0000 AUD | Free ($0.00 AI Voice usage) |
hybrid.sip.crazytel.net.au) linked to your Voice Profile using that trunk's Caller ID.Explore the interactive OpenAPI schema, copy ready-to-run curl examples, and inspect all 15 production REST endpoints.
Example 1: Place a Call with Just 2 Fields (destination + initial_text)
curl -X POST "https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls" \
-H "X-Crazytel-Api-Key: $CRAZYTEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"destination": "0491570156",
"initial_text": "Hi, I'\''m calling to confirm your 10am appointment tomorrow."
}'
Example 2: Smart Yes/No Goodbye & Automatic Voicemail Drop
curl -X POST "https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls" \
-H "X-Crazytel-Api-Key: $CRAZYTEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"destination": "0491570156",
"initial_text": "Hi Sarah, I'\''m calling from Acme Solar to confirm if someone will be home at 10am tomorrow?",
"closing_message_mode": "conditional",
"closing_text_affirmative": "Wonderful, thank you for confirming! We'\''ve locked in 10am tomorrow. Have a great day, goodbye!",
"closing_text_negative": "No problem at all, I'\''ll let our scheduling team know so we can text you alternative times. Goodbye!",
"closing_text": "Thank you, I'\''ve passed your message along to our team. Have a great day, goodbye!",
"voicemail_detection_enabled": true,
"voicemail_action": "leave_voicemail",
"voicemail_text": "Hi Sarah, this is Acme Solar calling to confirm your 10am installation tomorrow. Please call us back on 1300 000 111."
}'
Example 3: Full AI Phone Agent (grok_voice_agent + append_instructions) - Keep your company's base instructions saved in the portal (default_instructions) and pass call-specific details via append_instructions:
curl -X POST "https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls" \
-H "X-Crazytel-Api-Key: $CRAZYTEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"destination": "0491570156",
"mode": "grok_voice_agent",
"append_instructions": "You are calling John Smith to confirm his 2pm solar roof inspection tomorrow at 42 Wallaby Way and ask if there is a gate code or dog in the yard."
}'
Example 4: Live Interactive Actions (/say, /refer, /dtmf, /hangup) - When mode is "interactive", the call remains open (status: "awaiting_next_turn") so your bot can issue follow-up commands:
# Speak the next turn and listen for their reply
POST https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/say
{
"text": "2pm tomorrow works great! Is there a gate code our technician should know about?",
"listen_for_reply": true
}
# Transfer the live call to staff extension 101 via Hybrid SIP REFER
POST https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/refer
{
"target": "101",
"announcement_text": "One moment please while I transfer you to our scheduling team."
}
# Press keypad digits on an IVR menu
POST https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/dtmf
{
"digits": "1",
"listen_for_reply": true
}
# Speak a custom closing message and hang up cleanly
POST https://api.crazytel.io/api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/hangup
{
"closing_text": "Perfect, I've noted that down and locked in 2pm tomorrow! Goodbye!"
}
https://api.crazytel.io)| Method | Endpoint Path | Description |
|---|---|---|
| POST | /api/v1/crazyai-voice-api/{profile_uuid}/calls |
Place an outbound AI Voice call using profile_uuid (single_turn, interactive, or grok_voice_agent) and return callee_transcript, analysis, turns, session, and AUD billing. |
| GET | /api/v1/crazyai-voice-api/{profile_uuid}/calls |
List call history, transcripts, and AUD billing for profile_uuid (?limit=50&offset=0). |
| GET | /api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid} |
Get full call details, turn-by-turn transcript (turns[]), Crazyswitch AI AMD analysis, and AUD billing. |
| POST | /api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/say |
Speak next turn (text, listen_for_reply, end_call_after) or leave a custom voicemail after the beep (wait_for_beep: true). |
| POST | /api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/refer |
Transfer an active call via Hybrid SIP REFER (target: extension 101, park 700, phone number, or sip: URI; optional announcement_text). |
| POST | /api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/dtmf |
Send DTMF keypad tones (digits: 0-9, *, #, A-D, , pause) and transcribe the next prompt. |
| POST | /api/v1/crazyai-voice-api/{profile_uuid}/calls/{call_uuid}/hangup |
Optionally speak closing_text and cleanly hang up (SIP BYE). |
| GET | /api/v1/crazyai-voice-api/profiles |
List all CrazyAI Voice API profiles (profile_uuid) with live AUD pricing. |
| POST | /api/v1/crazyai-voice-api/profiles |
Create a reusable Voice Profile linked to a PAYG Hybrid SIP Trunk (only name + hybrid_username or trunk_id required). |
| GET | /api/v1/crazyai-voice-api/profiles/{profile_uuid} |
Get a single Voice Profile configuration and its ready-to-use call_endpoint. |
| PUT | /api/v1/crazyai-voice-api/profiles/{profile_uuid} |
Partially update a Voice Profile. |
| DELETE | /api/v1/crazyai-voice-api/profiles/{profile_uuid} |
Permanently delete a Voice Profile and its associated call records. |
| GET | /api/v1/crazyai-voice-api/voices |
List all 28 built-in AI voices, 21 language codes (including auto), expressive speech tags, and live AUD pricing. |
| POST | /api/v1/crazyai-voice-api/voices/preview |
Synthesize up to 500 characters into a 16 kHz mono WAV preview clip (audio_base64). |
| GET | /api/v1/crazyai-voice-api/pricing |
Get live retail AUD pricing for TTS playback, streaming/REST STT, Full AI Phone Agent, and Crazyswitch AI AMD. |
Crazytel builds for responsible, compliant Australian business communications:
Outbound calls require a Crazytel Hybrid SIP Trunk on a Pay-As-You-Go (PAYG) plan.
Any sales, telemarketing, or promotional calling must comply with the Do Not Call Register Act 2006 (Cth) and wash contact lists against donotcall.gov.au unless valid consent is held.
Calls must adhere to ACMA rules, permitted calling hours in the recipient's local Australian time zone, and upfront business identification in initial_text.
Outbound calls display your owned/verified Hybrid SIP Trunk Caller ID (13/1300/1800 inbound-only numbers cannot be used as outbound CLI), with a one-click profile toggle to prepend "Please note this call is being recorded" whenever required.
Everything you need to know about CrazyAI Voice API setup, call modes, Crazyswitch AI AMD, and PAYG Hybrid SIP Trunk billing.
callee_transcript).
profile_uuid. Your app or AI bot simply sends a standard HTTPS POST request with destination and initial_text, and receives a JSON response containing the transcript, intent analysis, and billing summary.
leave_voicemail), hang up immediately at $0.00 TTS cost (terminate), report voicemail_detected back to your bot (report_only), or continue normally (nothing).
conditional) (either in the portal or per call in the API). If the customer confirms, the AI speaks your affirmative goodbye; if they decline, it speaks your negative goodbye.
interactive mode (with Allow Call Transfers to Staff enabled), your bot can call POST .../refer to transfer the live call to an internal Hybrid extension (such as 101), a call park slot (700), or an external phone number, optionally speaking a transfer announcement first.
keyterms), and standing agent instructions (default_instructions). On any individual API call, you can omit those fields to use your profile defaults, override them for that specific call, or combine them (for example, sending append_instructions or extra_keyterms to add call-specific context to your saved profile defaults).
curl snippets inside the portal at Products → CrazyAI Voice API → API Management (/crazyai-voice-api/api-management).
Create your Voice Profile in 30 seconds, preview 28 lifelike voices, and place your first outbound AI call with just destination and initial_text.
Ask about Crazytel products or setup
Ask CrazyAI
Questions about Crazytel products or published setup steps. Include your device make and model for device questions, and do not include passwords or payment details.
By chatting here, you agree we and authorized partners may process, monitor, and record this chat and your data in line with Privacy Policy.
