# Optional real-time media stack — enables agent REPLY-AUDIO in live calls (#64).
#
# Install where calls with agent voice are used (HART OS desktop / a central SFU):
#     pip install -r requirements-voice.txt
#
# Deliberately kept OUT of the base requirements.txt:
#   - `livekit` is a ~11MB native package (bundles the Rust liblivekit_ffi). Not
#     every deploy needs it, and the code degrades to a logged no-op without it:
#     integrations/social/livekit_audio_publisher.py and agent_voice_bridge.py
#     are lib-gated on HAS_LIVEKIT_RTC, so a base install is unaffected.
#   - It requires protobuf>=4.25 + aiofiles>=24. The base file still pins those
#     lower (protobuf==4.23.3 / aiofiles==23.2.1) for the legacy chromadb stack;
#     installing this file resolves aiofiles>=24 in the voice env only (protobuf
#     is already 5.x in practice). Folding livekit into the base file would make
#     `pip install -r requirements.txt` fail the resolver on the protobuf pin.
#
# Pairs with the already-present livekit-api / livekit-protocol (token signing +
# the self-hosted SFU supervisor); this adds the realtime media half (livekit.rtc)
# that the AgentVoiceBridge publishes the agent's TTS frames through.
livekit==1.1.10