Skip to content

README.md

Latest commit

 

History

History
56 lines (39 loc) · 1.93 KB

README.md

File metadata and controls

56 lines (39 loc) · 1.93 KB
Raw

ChatAppIPC

Foundry Local runs chat and speech models directly on Android devices. ChatAppIPC demonstrates streaming chat and live voice input while inference runs in the Foundry Local service app.

What It Demonstrates

  1. Connect to the service with FoundryLocalManager.create(...).
  2. Download and load a chat model through the shared coroutine API.
  3. Stream chat tokens with completeChatStreaming(...).
  4. Download and load the Nemotron speech model.
  5. Capture microphone audio and transcribe it with an AudioStreamSession.
  6. Clear stale model and client handles when the service disconnects.

Good to know: This sample uses only com.microsoft.foundrylocal.api.*. It does not use the older IPC callback API, FLResult, or AIDL callback stubs.

Build and Run

  1. Download foundry-local-ipc-sdk-<version>.aar from the matching GitHub Release, verify its published SHA-256 hash, and place it in:

    examples/ipc/ChatAppIPC/libs/

  2. Install the Foundry Local service app from Google Play.

  3. Build the app from the repository root:

    ./gradlew :ChatAppIPC:assembleDebug

  4. Connect an Android device and install the app:

    ./gradlew :ChatAppIPC:installDebug

  5. Open Chat from the device launcher.

Good to know: Keep only one IPC AAR in the sample's libs/ directory. Gradle loads every AAR in that directory, and multiple versions cause duplicate-class errors.

Use Voice Input

  1. Complete the chat model setup.
  2. Wait for the Nemotron speech model to download and load.
  3. Grant microphone permission.
  4. Tap the microphone button and speak.
  5. Tap stop to copy the final transcript into the chat input.

The chat model is qwen2.5-coder-0.5b-instruct-generic-cpu:4. Voice input uses nemotron-speech-streaming-en-0.6b-generic-cpu:3, which requires an additional model download.