# 1.3.0

## AI Assistants

### Features

#### Data injection

* Inject data when a conversation starts. Each key becomes a dynamic variable, resolved with `{key}` in the Assistant's instructions, first message, and other templated fields. See [Data injection](/docs/data-injection).
* Inject data through UUI on [SIP Channels](/docs/data-injection#injecting-data-via-sip-channels).
* Set `injectedData` in the `POST /calls/:channelId` request payload for [WebRTC Channels](/docs/data-injection#injecting-data-via-webrtc-channels).

### Improvements

#### Startup time latency

* Reduced the time between connecting to a conversation and the Assistant being ready by 50%.

#### Set fixed audio output samplerate

* Set audio output to 48 kHz to match the native Opus sample rate and remove a resampling step before WebRTC encoding.

#### Ultravox support

* Added the Ultravox realtime models Gemma3 27B Preview, Llama 3.3 70B, and Ultravox Realtime.
* Added Inworld and ElevenLabs voices for Ultravox.

#### Gradium support

* Added the default Gradium STT model.
* Added the default Gradium TTS model and voices.

#### GPT Realtime 1.5 & Mini support

* Added the `gpt-realtime-1.5` and `gpt-realtime-mini` models.

## Calls API

### Features

#### Data injection

* `POST /calls/:channelId` accepts an `injectedData` object. See [WebRTC data injection](/docs/data-injection#injecting-data-via-webrtc-channels).
