Skip to content

Status and support

This page says what is available today and what is planned. PyPI’s voqalize-agent-sdk page and the @voqalize/client-transport CHANGELOG are the upstream records for the SDK surfaces.

Voqalize fits an app that already has a screen and users, and wants them to talk to an agent that acts on that screen: a store, a bank’s support flow, a booking form, an internal console. You keep the agent’s logic, model and data in your own backend; Voqalize runs the voice. Examples shows the patterns.

Today it does not place or answer phone (PSTN) calls, it does not host your agent’s model, and it runs in India only.

The published Python SDK is voqalize-agent-sdk==0.15.0, for Python 3.12 or later. Pin the version you build against: releases before 1.0 may change the brain API.

ExtraBrain it ships
geminiGeminiBrain (recommended), and GeminiInteractionsBrain (experimental; see the shipped adapters)
openaiOpenAIBrain
litellmLiteLLMBrain

Another framework or language connects through the published protobuf wire, checked by the conformance harness. Use another agent framework shows how.

The page side is your client on @voqalize/client-transport@0.4.1, which handles the media. See Client libraries.

EnvironmentClientVoqalize example
Web@pipecat-ai/client-js with @voqalize/client-transportYes
React web app@pipecat-ai/client-react with @voqalize/client-transportYes

Voqalize is pipecat-compatible, and a native app uses pipecat’s own client SDKs.

EnvironmentClientVoqalize example
React NativePipecat React Native with SmallWebRTCNot yet
Native iOSPipecat Swift with SmallWebRTCNot yet
Native AndroidPipecat Kotlin with SmallWebRTCNot yet

Every environment on this page is supported. A native app starts sessions from your backend with an sk_ key, because a pk_ key works only from a browser page (Client libraries).

Both connections are supported and run the same brain code. An inbound WebSocket route in your app is the production default. The outbound Cortex relay is the local-development default, and suits serverless and egress-only networks. See Deploy the brain.

There is one Voqalize environment, app.voqalize.com, and you use it for local development and testing as well as for production. Sessions are processed and stored in India today. A United States region is planned and cannot be selected yet.

A session lasts at most one hour. Every other limit, and what happens at it, is in Usage and limits; included minutes and storage are set by your plan (pricing).

DataStored by Voqalize
Session record and initYes
Lifecycle and wire eventsYes
TranscriptsYes, as part of the wire-event record
Voqalize logsYes
Audio, and camera and screen videoOnly when the session records; video only for a lane the page turned on
Your brain’s model history and logsNo; they stay in your environment

Stored session data, recordings included, is retained for 30 days. Send identifiers rather than personal data in init, and resolve them against your own store.

Voqalize is priced per conversation minute. The minute includes speech recognition, speech synthesis, turn-taking, WebRTC, the avatar and observability. Your brain’s model usage stays on your own model-provider account.

Plans, included minutes, the free signup credit, support and uptime commitments are on the pricing page.