voice-agents

A voice agent demo simulating "Acme Bank" - a banking assistant for credit cards, loans, and investments.

• RedHat AI Platform
• Open Source models
  • Speech to text using whisper LLM
  • Text to speech using higgs-audio LLM
  • Supervisor using llama-4-scout LLM (maas)
• Agent handoffs using Langchain + Langgraph
• Next.js web ui using websockets
• Python websocket server backend
• (Optional) MLFlow tracing

Helm Install

Deploy the voice agent stack (backend, frontend, MLflow) to OpenShift/Kubernetes:

helm upgrade --install ai-voice-agent ./ai-voice-agent/deploy/chart \
    --set backend.env.BASE_URL="<LLM_ENDPOINT_URL>/v1" \
    --set backend.env.MODEL_NAME="<LLM_MODEL_NAME>" \
    --set backend.env.TTS_URL="<TTS_ENDPOINT_URL>/v1" \
    --set backend.env.TTS_MODEL="<TTS_MODEL_NAME>" \
    --set backend.env.TTS_VOICE="belinda" \
    --set backend.env.STT_URL="<STT_ENDPOINT_URL>/v1/audio/transcriptions" \
    --set backend.env.STT_MODEL="whisper" \
    --set backend.secret.API_KEY="<YOUR_API_KEY>" \
    --set backend.secret.STT_TOKEN="<YOUR_STT_TOKEN>"

To disable MLflow tracing:

    --set mlflow.enabled=false

To point at an external MLflow instance instead of the in-cluster one:

    --set mlflow.enabled=false \
    --set backend.env.MLFLOW_TRACKING_URI="http://your-mlflow-host:5500"