Private RT melia 1 docs

docs/private/next-gen-model-realtime.mdx

@@ -0,0 +1,88 @@
+---
+description: Get started with our next-generation model for Realtime transcription
+---
+
+# Next-generation Model – Realtime
+
+## Introduction
+We are offering existing customers an early preview of our next-generation model, `melia-1`, for Realtime transcription. This is the Realtime companion to the [Batch preview](./next-gen-model).
+
+:::warning
+This is an early preview for evaluation only, and is not intended for production use. Only a subset of Realtime features is currently supported (see [Capabilities](#capabilities)). Remaining features will be added over time.
+:::
+
+This page describes how to access the preview and what to expect in terms of capabilities.
+
+## Getting started
+You can start right away using your existing API keys, no new credentials needed. If you don't have an API key, head to the [Portal](https://portal.speechmatics.com) to get one.
+
+## API reference
+The API schema is consistent with the [V2 Realtime WebSocket API](https://docs.speechmatics.com/api-ref/realtime-transcription-websocket) with some additions and limitations described below.
+
+### Supported endpoint
+The preview runs on a dedicated endpoint, separate from the production Realtime service:
+
+| Region            | Endpoint                                 |
+|-------------------|------------------------------------------|
+| Preview (EU & US) | **wss://preview.rt.speechmatics.com/v2** |
+
+The preview is deployed in France (EU) and Oregon (USA).
+
+### Minimal example
+Example minimal `StartRecognition` message:
+
+```json
+{
+  "message": "StartRecognition",
+  "transcription_config": {
+    // highlight-start
+    "language": "multi",
+    "model": "melia-1"
+    // highlight-end
+  }
+}
+```
+
+### LanguageInfo message
+A new `LanguageInfo` message is sent from the transcriber to the client the first time a new language code is detected in the audio. It is emitted **once per language code**, immediately before that language first appears in an `AddTranscript` (Final) or `AddPartialTranscript` (Partial) message.
+
+Use it to configure how you process and display the transcript — for example, word spacing and text direction for right-to-left scripts.
+
+```json
+{
+  "message": "LanguageInfo",
+  "language": "ar",
+  "word_delimiter": " ",
+  "writing_direction": "right-to-left",
+  "partial": true
+}
+```
+
+- `language` (String): The detected language code.
+- `word_delimiter` (String): The delimiter to use between words when reconstructing the transcript for this language.
+- `writing_direction` (String): Either `left-to-right` or `right-to-left`.
+- `partial` (Boolean): Optional. When present, this is `true`, indicating the message was triggered by a Partial transcript.
+
+## Capabilities
+
+| Feature                                     | Current capability          | Future plans                           |
+|---------------------------------------------|-----------------------------|----------------------------------------|
+| Partials                                    | ✅ Matches current behaviour |                                        |
+| Channel diarization                         | ✅ Matches current behaviour |                                        |
+| Output locale                               | ✅ Matches current behaviour |                                        |
+| Word timings                                | ✅ Matches current behaviour |                                        |
+| Language labelling                          | ✅ Word-level                |                                        |
+| Speaker diarization                         | ⚠️ Not yet                  | Prioritised according to customer need |
+| Speaker identification                      | ⚠️ Not yet                  | Prioritised according to customer need |
+| Language hints                              | ⚠️ Not yet                  | Prioritised according to customer need |
+| `max_delay` / `max_delay_mode`              | ⚠️ Not yet                  | Under consideration                    |
+| Mid-session config (`SetRecognitionConfig`) | ⚠️ Not yet                  | Prioritised according to customer need |
+| End of Utterance (EOU / FEOU)               | ⚠️ Not yet                  | Prioritised according to customer need |
+| Custom dictionary                           | ⚠️ Not yet                  | Prioritised according to customer need |
+| Alphanumerics                               | ⚠️ Not yet                  | Prioritised according to customer need |
+| Entities / punctuation overrides            | ⚠️ Not yet                  | Prioritised according to customer need |
+| Disfluency removal                          | ⚠️ Not yet                  | Prioritised according to customer need |
+| Profanity tagging                           | ⚠️ Not yet                  | Prioritised according to customer need |
+| Audio volume filtering                      | ⚠️ Not yet                  | Prioritised according to customer need |
+| Translation                                 | ⚠️ Not yet                  | Prioritised according to customer need |
+| Audio events                                | ⚠️ Not yet                  | Prioritised according to customer need |

docs/private/next-gen-model.mdx

@@ -6,7 +6,7 @@ description: Get started with our next-generation model
 ## Introduction
 We are excited to offer early access to our next-generation model, codenamed Omni. This new architecture is designed to provide improved multilingual capabilities, faster turnaround time and improved alphanumeric accuracy across all supported languages.
 
-Currently available for Batch transcription, the Omni model will be made available for Realtime transcription in Q3 2026.
+Currently available for Batch transcription and Realtime transcription — see [Next-generation Model – Realtime](./next-gen-model-realtime).
 
 This page will guide you through how to access the new model, and what to expect in terms of capabilities.