Latency
Conversations without the wait
~0.9 sp50
1.2 sp95
Measured from end of turn to first agent audio on the fast profile.
Voice agent API
Build voice agents that respond naturally. Not expensively.
Real-time conversational AI with natural voices, custom agents, memory, and tool calling. One simple API. No complex audio infrastructure. Pay per turn, not per minute.
~0.9 smedian reply (p50)
$0.003per turn, Fast tier
~50stock voices + cloning
Live demo
Talk to a Ciabatta agent
A short call with a sample agent. Uses your microphone; nothing is recorded or stored.
Ready
Try asking:
"What can you do?"
"How fast can you reply?"
"Interrupt me mid-sentence"
Pricing
Pay for conversation, not silence
$0.003per turn (Fast)
$0.13per idle hour
You are billed per agent turn, so listening time is almost free. Each turn includes 6,000 prompt tokens, 400 output tokens, 400 speech characters and 45 s of connected time.
Control
Your agents. Your rules.
- Choose from ~50 voices or clone your own
- Pick the model tier per agent
- Webhook, client and built-in tools
- Per-customer memory and summaries
- Custom personality and instructions
Agent Studio
Configure an agent in a few clicks
Set the voice, model, tools, memory and personality from the dashboard, or do it all through the API. Changes are versioned, so you can test safely.
Warm, friendly Preview
S FastML
You are a concise, upbeat support agent for a bakery. Confirm orders before booking.
check_hoursbook_slottransfer_call
Remember each customer
Pricing
Your cost advantage, calculated
Ciabatta-
Example competitor-
Estimated savings-
Illustrative monthly estimate at 3.5 turns per minute, using the Ciabatta per-turn price for the selected tier (- per turn, - per minute) against an illustrative competitor price of $0.10 per minute. Actual costs vary with turns, allowances, model choices and surcharges.
| Model tier | Per turn | ~Per minute |
|---|---|---|
| S (fastest, default) | $0.0030 | ~$0.011 |
| M | $0.0045 | ~$0.016 |
| L | $0.0065 | ~$0.023 |
Small surcharges apply only beyond the per-turn allowances, and every turn reports its usage. Prepaid credits, no minimums.
See the same usage at other platforms
Competitor costs use the typical all-in per-minute ranges below (light bar = high end of the range).
| Platform | Typical all-in price | Billing |
|---|---|---|
| Ciabatta | $0.011-0.037/min | Per turn; idle listening $0.13/h |
| Vapi | $0.10-0.30/min | Per minute + providers |
| Retell AI | $0.10-0.20/min | Per minute |
| ElevenLabs Agents | ~$0.08-0.10/min | Per minute, plan-bound |
| Bland | ~$0.09/min | Per minute |
| OpenAI Realtime | ~$0.10-0.30/min | Per audio token |
Competitor prices are approximate public list prices as of October 2026 and vary by plan and provider choice.
Developers
Ship in minutes
Create a session on your server, hand the WebSocket URL and token to your web or mobile app, and you have a talking agent.
// npm install @getciabatta/sdk
import { Ciabatta } from "@getciabatta/sdk";
const ciabatta = new Ciabatta({ apiKey: process.env.CIABATTA_API_KEY });
const session = await ciabatta.sessions.create("support", { customer_id: "cust_123" });
// hand session.ws_url + session.token to your web or mobile appBuilt for real conversations
Echo cancellation
Server-side, fed with the exact audio the caller hears.
Barge-in
Talk over the agent and it stops immediately.
Tools
Webhook, client and built-in tools.
Memory
Per-customer conversations and summaries, with your retention settings.
Resilient connections
Dropped connections resume where they left off, with no lost or repeated audio.
Natural voices
Nearly 50 stock voices, plus cloning from a short recording.