diff --git a/docs/realtime-input-audio-stream-end-support.md b/docs/realtime-input-audio-stream-end-support.md
new file mode 100644
index 0000000000..81bba522a6
--- /dev/null
+++ b/docs/realtime-input-audio-stream-end-support.md
@@ -0,0 +1,12 @@
+# Realtime Input: audio_stream_end Support
+
+ADK supports sending the `audio_stream_end` signal in realtime streaming inputs to the Gemini Live API.
+
+## Overview
+
+When Voice Activity Detection (VAD) is active, the Gemini Live API requires an explicit `audio_stream_end` signal to flush cached audio buffers upon completion of user speech.
+
+## Shipped Changes
+
+- **`LiveClientRealtimeInput` Handling**: `GeminiLlmConnection`, `BaseLlmFlow`, and `LiveRequestQueue` accept generic `LiveClientRealtimeInput` messages with `audio_stream_end` configured.
+- **Buffer Flushing**: Sending `audio_stream_end=True` notifies the backend to process and flush buffered audio rather than waiting for subsequent chunks.