A multi-tenant voice calling agent built with Python + Flask. Create an agent for each client, upload their knowledge base (CSV / JSON / paste), white-label it, and install it on their website with a single script tag.
Mic is paused while the agent speaks - the agent can no longer hear itself and loop.
STOP button + Esc key kill speech and mic instantly, with clean teardown - no zombie audio.
Replies are split into sentences and queued, avoiding Chrome's 15-second speech cutoff.
Noisy environment? Switch to push-to-talk, or enable confirm-before-send for accents.
Per-client CSV/JSON/TXT upload with FAQ support. Import the full Tukaram catalog as a template.
Client name, agent name, colors, greeting, contact info - fully branded per tenant.
Names, phones and interests auto-captured. CSV export + read-only client report link.
Runs free out of the box. Add OpenAI, Claude, Groq or Ollama per client for smoother answers - with automatic fallback.
Copy an agent, export a bundle, or deploy a fresh copy of this app with different knowledge.
Abuse protection per IP and per call, built in.