IA

Realtime API

API service

by Inworld AI

Speech-to-speech voice-agent API in which one WebSocket session runs Inworld speech recognition with voice profiling, a chosen LLM from 100+ with tool calling and turn detection, and Realtime TTS-2 voice output.

Find alternatives to Realtime API

What it does

• Voice agent
• Speech to text
• Text to speech

Features 6

• Semantic turn detectionSemantic VAD reads the transcript to decide when the user has finished, with adjustable eagerness.
• InterruptionsAudio stops when the user talks over a reply, and history keeps only what the user heard.
• Tool callingFunctions are declared at the start or added mid-session and the agent speaks the result in the same turn.
• Mid-session settings changesA session.update changes the LLM, voice or tools for the next reply.
• OpenAI Realtime compatibilityEvents and session config match the OpenAI Realtime API, so migrating means a new endpoint and key.
• Browser and phone audioConnects over WebSocket or WebRTC and accepts G.711 8 kHz phone audio from Twilio.

Sources

Pricing

Pricing page · 11 Oct 2026

"Realtime API access" from the On-Demand plan; usage is metered at each tier's TTS, STT and LLM rates ("LLMs At cost") and deducted from plan credits. Pricing page plan cards: On-Demand ("Start free", pay as you go), Creator $25/mo, Builder $100/mo, Developer $300/mo, Growth $1,500/mo, each granting equal monthly credits, and Enterprise "Custom pricing, limits & terms" with "contact sales". No free allowance is stated for Realtime API usage.

Description

Official documentation · 11 Oct 2026

"One WebSocket call runs STT voice profiling, LLM steering, and Realtime TTS-2 expressive output. Sub-second latency, 100+ LLMs"; configuration tabs "LLM Choice | TTS Voice | Tool Calling | Turn Detection".

What it does

Official documentation · 11 Oct 2026

"Controllable speech-to-speech that understands, reasons, and interacts".

Sold within

Pricing page · 11 Oct 2026

Self-serve plans; no parent offering.

Name

Official documentation · 11 Oct 2026

Page heading "Realtime API".

URL

Official documentation · 11 Oct 2026

Product page "Realtime API: Expressive Speech-to-Speech with Realtime TTS-2".

Deployment

Official documentation · 11 Oct 2026

Hosted WebSocket API.

Interfaces

Official documentation · 11 Oct 2026

"Realtime API Quickstart: Build a voice agent with WebSocket streaming"; OpenAI Realtime migration guide.

Modalities

Official documentation · 11 Oct 2026

"output_modalities": ["audio", "text"] with audio input.

Intended for

Official documentation · 11 Oct 2026

Developers building voice agents for consumer applications.

Also from Inworld AI

• Realtime TTS API service

  Text-to-speech API (Realtime TTS-2 and TTS-2 Flash) with voice cloning and voice design, billed per million characters.

• Realtime STT API service

  Speech-to-text API (Inworld STT-1) that transcribes streaming or complete audio in 30+ languages with turn detection, returning voice-profile signals for emotion, vocal style, accent, age and pitch alongside the transcript.

• Realtime Router Platform

  OpenAI-compatible LLM gateway that sends each request to one of 100+ models from OpenAI, Anthropic, Google and others by rules on user attributes, context, A/B splits or cost, with failover and provider rates passed through at zero markup.

• Inworld Compute Infrastructure service

  Dedicated NVIDIA B200 GPUs for high-volume realtime workloads such as LLM inference, dedicated TTS serving and custom model hosting, with Kubernetes access or Inworld-managed deployment, and deployed models routable through Inworld Router.

• Inworld Inference Model API

  Inworld Inference provides API access to text, image, video and decision models.

Open Inworld AI in the directory

Something wrong here? Send a correction — quote this product id: inworld-realtime-api.