BoilerShelf Voice Chatbot

A full-stack boilerplate for building turn-based conversational voice AI assistants. Speak into the microphone, get a transcription, a context-aware LLM reply, and a synthesized voice response streamed back — all with zero external speech-API fees.

Core capabilities

• Local Speech-to-Text: whisper.cpp (base.en model) transcribes audio locally, no cloud STT costs.
• Local Text-to-Speech: Kokoro-82M (am_adam voice) synthesizes natural speech locally, no cloud TTS costs.
• Conversational memory: multi-turn history preserved per session/thread.
• Dual input: speak via microphone or type — the assistant responds in both text and audio.

Four tiers, two independent choices

Pick your database (MongoDB or PostgreSQL) and your auth level (NoAuth or WithAuth) — all four combinations ship the identical voice engine:

• NoAuth tiers: no login wall, single-tenant, ideal for prototyping and internal tools.
• WithAuth tiers: adds JWT authentication (bcrypt hashing, refresh cookies, email OTP verification, password reset), role-based access control, an admin dashboard, and audit logging.

This is a pure conversational voice agent — no document upload, no RAG, no vector database. If you need document-grounded AI instead, see the LangGraph RAG Agent product.