Voice AI Infrastructure
for Production.
Three engines on one platform: cascading, speech-to-speech and full duplex. Web, phone and SIP channels, MCP tools and full observability built in.
Three engines
Choose the architecture per agent. Phone numbers, tools, knowledge and call logs work the same on all three.
Talk to the agent live
Start a live conversation with our voice agent. No signup required.
Build it once, answer everywhere
Web voice, inbound and outbound calls and SIP trunks all reach the same agent, with the same prompt, tools and knowledge.
Configure it, then talk to it
Voice, model, prompt, tools and knowledge on one screen, with a test call next to it.
You are the front desk for Acme Dental. Greet {{caller_name}} by name and look up {{user_number}} before you ask anything.
It is {{current_time_America/New_York}}. If the office is closed, offer to book a callback.
Use book_appointment to schedule. Never quote prices.
Built for production, not just demos.
The voice AI platform designed for the call volume, failure modes, and compliance requirements of real production systems.
One API call to deploy
Define a prompt, point to a number, and your agent handles real calls. No telephony account to set up, no deployment pipeline to configure, no infrastructure YAML to write.
POST /agent/start
{
"prompt": "You are a sales agent",
"phone_number": "+1..."
}Full observability
Call transcripts, per-segment LLM latency traces, DTMF event logs, and failure replay. Post-call extraction pulls the fields you define per agent out of every call, through the API, dashboard and webhooks.
Telephony built in
Inbound DIDs, outbound caller ID, call routing, and SIP trunking — all managed inside Blowfish. Bring your own carrier or use ours.
Native MCP integrations
Connect to your CRM, scheduling system, EHR, or internal APIs via Model Context Protocol. Agents book appointments, update records, and trigger workflows mid-call.
Native speech-to-speech
First-class support for realtime models: OpenAI GPT Realtime (2, 2.1 and Mini), GPT-Live-1 and xAI Grok Voice Agent. Audio in, audio out, no transcription step. Cascading STT, LLM and TTS is there when you want to pick each stage.
Frequently Asked Questions
Still have questions? Contact Us
Tell us what you are building
Tell us what you are making. We will walk you through the engines, the API, and what it takes to run it in production.