💡 Target Audience: Engineering teams building cross-border products who are tired of stitching DeepL or Google Translate calls into every chat send.

The "translation as duct tape" problem

If you've built a multilingual chat app on SendBird, Stream, Twilio, or any general WebSocket pipe, the flow probably looks like this:

1. User types a message

2. You send it to your backend

3. Backend calls DeepL / Google Cloud Translate / OpenAI

4. Backend stores the translation

5. Backend pushes the translated copy back through the chat API

6. Chat API broadcasts to recipients

That's six hops. Each hop adds latency. Each hop is a billing line. Each hop is a failure point. And every recipient who reads in a different language multiplies the call count.

When you scale to thousands of concurrent rooms with mixed-language participants, the translation orchestration becomes the most expensive and fragile part of your system.

HyperBabel's approach: translation lives in the channel

HyperBabel collapses those six hops into one. When a sender publishes a message, the channel itself knows the recipients' preferred languages and runs translation inline — before the message lands in each recipient's inbox.

```javascript

// One call. Translation is part of the message contract.

await channel.sendMessage({

text: "Welcome to the launch event!",

translate_to: ['ko', 'ja', 'es', 'pt', 'de', 'fr']

});

// Each recipient receives the message in their preferred language,

// with the original always available via message.original_text.

```

The same primitive works for:

* 1:1 and group chat — every participant sees their own language

* Open rooms (community chat) — broadcasts to thousands of viewers, each rendered locally

* Video subtitles — speaker captions auto-translated for global audiences

* Live streaming chat — viewer comments in 100+ languages, translated in the same broadcast pipeline

Why "in the channel" matters technically

Three things change when translation isn't bolted on:

1. Single round-trip latency. No external translation API hop. Translation happens at the edge node closest to the sender, parallel to the message persistence write. End-to-end translated delivery typically lands in the same p50 <65ms / p95 <150ms budget as untranslated messages — because translation isn't sequential to delivery, it runs alongside it.

2. No double-billing. Competitors bill you for chat (per MAU or per message) AND for translation (per character through DeepL/Google). HyperBabel bills only the chat plan — translation characters are part of the included quota (500K chars/mo on Starter, 5M on Pro, 20M on Business; pricing.hyperbabel.com).

3. Consistent failure model. If your separate translation provider rate-limits you, your chat looks broken — messages arrive untranslated for some users while others see fine, no easy retry surface. With translation inline, the message either delivers translated or returns a single deterministic error you can handle in one place.

Code: subtitle translation in a live broadcast

The same translation primitive applies to live streaming. A broadcaster speaks Korean; viewers in 100+ countries see subtitles in their preferred language without writing any translation code:

```javascript

const stream = await hb.live.startBroadcast({

title: 'Product launch',

// Subtitle translation auto-derived from each viewer's locale

subtitle_translation: 'auto',

source_language: 'ko'

});

// Viewers automatically receive subtitles in their device locale.

// No per-viewer API call needed.

```

Quality and language coverage

HyperBabel's translation engine covers 100+ languages including all major global tongues plus regional variants (zh-CN vs zh-TW, pt-BR vs pt-PT, en-US vs en-GB). The model is tuned specifically for chat-style short-form text — better at handling slang, emoji-laden sentences, code-switching, and conversational fragments than general document-translation APIs.

For domain-specific terminology (medical, legal, gaming, finance), custom translation models will arrive in late 2026 — see the [public roadmap](https://hyperbabel.com/roadmap).

What this means for your roadmap

If your product has any of these on the horizon:

* Expanding to a non-English market

* Cross-language customer support

* Live shopping or events with global viewers

* B2B platforms where Asian, European, and American teams collaborate

…you should not be wiring DeepL into your chat send handler. Architecturally, translation belongs in the channel, not in your application code.