AI Voice Agents on Real Phone Lines

Three API calls to a working voice agent

• Provision a number — POST /v1/numbers/purchase gives you a real US phone number in seconds.
• Register your agent — POST /v1/agents with your AI endpoint URL and endpoint type (WebSocket, OpenAI Realtime, HTTP, or CAIZA).
• Assign the number — POST /v1/numbers/{id}/assign links the number to your agent. Inbound calls are now routed to your AI.

Connect any AI engine

• WebSocket — Stream raw μ-law 8kHz audio bidirectionally. Works with Deepgram, AssemblyAI, or any custom STT pipeline.
• OpenAI Realtime — Native integration with the OpenAI Realtime API. Route calls directly without an intermediary server.
• HTTP — Send audio chunks as POST requests. Your endpoint returns audio to play back.
• CAIZA — Kadeno's default AI engine. Drop-in conversational AI for agents that don't need custom models.

Pay only for what you use

When to use Kadeno for voice agents

• You already have an AI pipeline — If you're running Whisper, GPT-4o, or ElevenLabs yourself, Kadeno provides the phone line without bundling AI you don't need.
• You want model flexibility — Switch STT, LLM, or TTS providers without changing your phone number or API integration.
• Cost at scale — At 10,000 minutes/month, Kadeno telephony costs $100. Bundled AI voice platforms run $1,000–3,000+ for the same volume.
• No lock-in — Your call logic lives in your code. Kadeno is infrastructure, not a framework.