allgpt-co/QuickVoice

QuickVoice – Open‑source AI Phone‑Agent Platform

What it is – A self‑hostable stack that lets engineering teams build, run, and extend voice‑based AI agents (inbound receptionists, outbound sales bots, etc.). It bundles a marketing site, a customer console, an Express API, a Python LiveKit worker, telephony integrations (Twilio/Telnyx), knowledge‑base (Pinecone) support, billing (Stripe), and all the surrounding infrastructure (Postgres, Redis, S3, etc.) in a single monorepo.

Why it matters – Most voice‑agent services are offered as closed SaaS APIs (e.g., Retell). QuickVoice gives you the same product surface with full source‑level control, enabling privacy reviews, cost transparency, and custom extensions that a hosted service can’t provide.

Core Features (as described in the README)

Category	Highlights
Agent runtime	LiveKit‑based voice worker written in Python; per‑call configuration (prompt, voice, language, dynamic vars); supports STT (Deepgram, Sarvam), LLM (AWS Bedrock), TTS (ElevenLabs, Sarvam).
Knowledge & tools	Pinecone vector store for RAG; Google embeddings; MCP/Smithery tool bridge; ability to call external HTTP tools during a call, with safety‑guards.
Telephony	Twilio and Telnyx number purchase, inbound routing, outbound calls, batch campaigns, LiveKit trunk binding.
Console & API	Next.js customer console for orgs, agents, numbers, KBs, API keys, billing, settings; Express API with auth, permissions, Stripe integration, webhook handling.
Privacy & compliance	Default PII redaction, optional zero‑PII mode, encrypted secrets, URL allow‑list, automated retention jobs for transcripts, recordings, KB rows, tool logs.
Billing	Stripe integration with test‑mode wallet flow; supports subscription‑style usage and prepaid credit.
Development experience	One‑command `task up:dev` (Docker Compose, pnpm, Python venv) launches the full stack locally; extensive Taskfile scripts for DB migration, seeding, CI, linting, testing.
Extensibility	Monorepo (pnpm + Turborepo) makes it easy to fork, add new providers, modify agents, plug in custom MCP tools, or replace the voice catalog.

Architecture Snapshot

Visitor → task up:dev →
  ├─ apps/web      (marketing site)
  ├─ apps/console  (customer UI)
  ├─ apps/server   (Express API, auth, DB, Stripe, Twilio/Telnyx)
  └─ apps/ai       (Python LiveKit worker, RAG, tool execution)
        │
        └─ LiveKit ↔ Telephony (Twilio/Telnyx)

All services run locally with Postgres and Redis containers; production requires you to supply your own LiveKit, telephony, and cloud provider credentials.

Tech Stack (from the README)

Frontend: Next.js (React) + Tailwind CSS
API: Express (TypeScript) with Better Auth
Voice runtime: LiveKit Agents, Python workers, Silero VAD, multilingual turn detection, noise cancellation
Telephony: Twilio, Telnyx
AI providers: Configurable STT/TTS/LLM via LiveKit inference (default: Deepgram, AWS Bedrock, ElevenLabs)
Knowledge & tools: Pinecone vector DB, Google embeddings, MCP/Smithery bridge, custom HTTP tools
Data stores: PostgreSQL (Prisma ORM), Redis/BullMQ, S3‑compatible object storage
Billing: Stripe
Orchestration: Docker Compose, Go Task, pnpm (v9), Turborepo
Language runtimes: Node 20/22+, Python 3.12, Go (for Task)

Project Maturity

Status: Active development, no stable release yet (see ROADMAP.md and draft v0.1.0 notes).
License: MIT (permissive, allows commercial use).
Community: Public issue forms, contribution guidelines, governance docs, and a roadmap. Stars and community interest are encouraged but the project is not yet production‑ready out‑of‑the‑box.

Typical Use Cases

Inbound AI receptionists – answer calls, look up customer data, provide support.
Outbound campaigns – automated reminders, sales calls, collections, with outcome tracking.
Appointment scheduling – voice‑driven booking flows.
Knowledge‑backed agents – answer questions from uploaded documents or vector‑store knowledge.
Domain‑specific automation – healthcare triage, financial services verification, logistics coordination, real‑estate lead qualification, e‑commerce support.

Getting Started (quick summary)

Prerequisites – Linux/macOS (or WSL2 on Windows), Bash ≥ 4, Node 20/22+, pnpm 9, Python 3.12, Docker Compose v2, Go Task.
Run – task up:dev creates env files, starts Postgres & Redis, runs migrations, and launches all services.
Access – Console at http://localhost:3000, marketing site at http://localhost:3001, API health at http://localhost:5000/api/v1/health, AI worker health at http://localhost:5555/health.
Add real credentials – Populate .env.dev with LiveKit, Twilio/Telnyx, AWS, Deepgram, ElevenLabs, Stripe, etc., to make live calls work.

Bottom Line

QuickVoice is a comprehensive, open‑source alternative to proprietary voice‑agent platforms. It gives teams the ability to inspect, customize, and self‑host every part of the voice‑AI stack—from the web console down to the telephony provider—while providing out‑of‑the‑box support for RAG, tool execution, billing, and privacy controls. It’s still early‑stage, but the repository contains a full monorepo, detailed docs, and a clear path to production for teams willing to manage their own infrastructure.