Current status and supported environments

This page separates what is available today from what is planned. The Python package on PyPI and Pipecat’s client documentation are the upstream records for the SDK surfaces.

Brain SDK

The published Python brain SDK is voqalize-agent-sdk==0.11.0. It requires Python 3.12 or later. Pin the version you build against: releases before 1.0 may change the brain callbacks.

The gemini extra ships GeminiBrain and GeminiInteractionsBrain.

GeminiInteractionsBrain is experimental

Every demo brain runs on GeminiBrain, and the last one that did not moved across on 2026-09-08. The interactions adapter’s interruption path has open defects — a step interrupted before its first delta stays in the context forever, and text buffered during a step is discarded if a barge-in lands before the step closes. Build on GeminiBrain; this one is kept, and tested, for the properties it has that generate_content does not.

Other agent frameworks and languages connect through the published protobuf WebSocket wire. The English specification and the conformance harness are the compatibility path: implement the wire around the framework’s own text stream, then run the harness before connecting a live call. See bringing an agent you already have.

App clients

Voqalize uses Pipecat’s RTVI and SmallWebRTC client surfaces directly. In a browser, the media is @voqalize/client-transport@0.4.0, our open implementation of pipecat’s media manager; pipecat’s stock transport still works and is documented for a couple more releases. See client libraries.

Environment	Client	Voqalize example
Web	Pipecat JavaScript with @voqalize/client-transport	Yes
React web app	Pipecat React with @voqalize/client-transport	Yes
React Native	Pipecat React Native with SmallWebRTC	Not yet
Native iOS	Pipecat Swift with SmallWebRTC	Not yet
Native Android	Pipecat Kotlin with SmallWebRTC	Not yet

All of these are supported client environments. The Voqalize repository currently ships complete examples only for web. For a native client, use the official Pipecat client and SmallWebRTC guide; the session-connect response supplies the endpoint and session credential that its transport uses. The current connection guide shows the web implementation.

Where the brain runs

Both connection paths are supported and run the same brain:

• An inbound WebSocket route in your application is the default deployment. Voqalize opens one connection per session.
• An outbound Cortex relay is supported for egress-only environments and is the default local-development path because the brain needs no public tunnel.

Build toward inbound for production unless the environment cannot accept a WebSocket connection. See where the brain runs.

Configuration ownership

The agent record stores the brain URL and an optional recording default. It stores no voice, language, recognizer or idle configuration.

sessions.connect accepts tts, stt, idle and record immediately before the call starts. A browser holding a publishable pk_ may call that route and set the same values, with one recording rule: record: false is accepted and record: true is refused. A backend holding an sk_ may set either value.

During the call, the brain may update tts, stt and idle with await session.configure(...). Recording is fixed when the call starts and cannot change mid-session. Voice and language and recordings carry the full rules.

Regions and availability

Calls are processed and stored in India today. A United States region is planned and is not available for selection yet.

The MCP server is rate limited per tenant; a refused request is a 429 with a Retry-After header. Concurrent sessions, included minutes and storage are set by your plan; pricing lists each plan’s allowance. Talk to us before a launch that needs more than a plan carries.

Stored session data

Data	Stored by Voqalize
Session record and init	Yes
Lifecycle and wire events	Yes
Transcripts	Yes, as part of the wire-event record
Voqalize logs	Yes
Audio, and camera and screen video	Only when recording is enabled for that session; video only for a lane the page turned on
Brain model history and brain logs	No; these remain 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.

Pricing and support

Voqalize is priced per conversation minute. The minute includes speech recognition, speech synthesis, endpointing, WebRTC, the avatar and observability. Your brain owns its model, so its LLM usage remains on your model-provider account and outside the Voqalize minute.

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

Read next

• A session, end to end — the complete application path.
• Quickstart — build the first web call.
• Usage and limits — the counters for your tenant.
• The wire — implement and verify a brain in another language.