Skip to content

v1.1.0 — Agora Conversational AI v2.12

Latest

Choose a tag to compare

@anandwana001 anandwana001 released this 15 Sep 04:32

Integrates features from Agora Conversational AI Engine v2.12.

Added

  • Android text controls: Ask Ada, Read aloud, and Queue instead of interrupting.
  • Backend think and speak endpoints with session checks and input validation.
  • Deferred instructions using append for listening, thinking, and speaking states.
  • Generated filler phrases after a 1.5-second wait, with static fallback phrases.
  • Optional getProjectGuidance custom tool for the public setup and troubleshooting guides.
  • Request-level tests for ASR serialization, custom tools, speech, and instructions.

Changed

  • Uses structured turn_detection.config and explicit speech-triggered interruption.
  • Pins the Python Agora Agents SDK to 2.8.1; retains Android RTC 4.6.4 and RTM 2.3.0.
  • App version is 1.1.0 (version code 2); backend default version is 1.1.0.

Fixed

  • Rejects placeholder App ID and App Certificate values at backend startup, instead
    of returning an unhandled HTTP 500 during token generation.
  • Corrects setup guidance to keep Agora credentials in server/.env.local.

Upgrade

  1. Install server/requirements-dev.txt in a Python 3.10+ virtual environment.
  2. Keep your real Agora App ID and App Certificate in server/.env.local.
  3. Optionally set PUBLIC_BASE_URL to the backend's public HTTPS base URL to enable
    project guidance. Deploy the repository's docs directory with the server.
  4. Restart the backend, configure Android with server/configure-android.sh, and
    rebuild/reinstall the app. Start a new conversation to apply agent settings.

Queued instructions begin a new turn after the current LLM output finishes;
this can precede completion of audio playback. Queued direct speech waits for
current speech. These features use the existing Android-to-backend REST path.

Validation

  • 10 Python backend tests and 26 Android unit tests pass.
  • Android debug build and lint pass.
  • Live emulator testing confirmed session start/end, typed instructions, and
    direct speech requests; the user confirmed the app works.
  • Generated-filler timing and LLM-driven guidance selection have request-level
    coverage but have not been independently verified in a live conversation.

Downloads

Use GitHub’s source archives below or clone tag v1.1.0. Configure your own Python backend and public HTTPS URL, then build the Android app. This release does not include an APK tied to a temporary development tunnel.