multimodal-voice-api-proxy 0.1.0

Privacy-first API proxy routing voice+vision to local models with cloud fallback

Multimodal Voice API Proxy

A privacy-first API proxy that intelligently routes voice and vision requests to local models with automatic cloud fallback.

What is this?

This proxy lets you run AI transcription (Whisper) and vision analysis (LLaVA) locally for privacy and cost savings, while automatically falling back to cloud APIs (OpenAI, Anthropic) when local resources are unavailable. It includes smart caching, usage metering, cost tracking, and multi-tenant API key management—perfect for developers building voice-enabled applications who want local-first privacy with cloud reliability as a safety net.

Features

• Intelligent routing: Local-first processing with automatic cloud fallback
• Privacy-focused: Audio and images processed on your hardware by default
• Smart caching: Redis-backed caching prevents reprocessing identical inputs
• Cost tracking: Real-time dashboard showing savings vs. cloud-only approach
• Multi-tenant: API key management with per-key usage limits and metrics
• OpenAPI-compatible: Drop-in replacement for OpenAI/Anthropic endpoints
• Production-ready: Rate limiting, monitoring middleware, and async processing
• Easy deployment: Docker Compose for local, one-click configs for Railway/Fly.io

Quick Start

Prerequisites

• Docker and Docker Compose
• 8GB+ RAM (for local models)
• GPU recommended but optional

Installation

1. Clone and configure

git clone <repository-url>
cd multimodal-voice-api-proxy
cp .env.example .env

2. Edit .env with your settings

# Required for cloud fallback
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

# Database
DATABASE_URL=postgresql://user:pass@db:5432/proxy

# Redis cache
REDIS_URL=redis://redis:6379/0

3. Launch with Docker Compose

docker-compose up -d

4. Run migrations

docker-compose exec api alembic upgrade head

5. Create your first API key

curl -X POST http://localhost:8000/keys \
  -H "Content-Type: application/json" \
  -d '{"name": "My App", "rate_limit": 100}'

The API will be available at http://localhost:8000

Usage

Audio Transcription

curl -X POST http://localhost:8000/v1/audio/transcriptions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F file=@audio.mp3 \
  -F model=whisper-1

Vision Analysis

curl -X POST http://localhost:8000/v1/vision/analyze \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image_url": "https://example.com/image.jpg",
    "prompt": "Describe this image"
  }'

Check Usage Stats

curl http://localhost:8000/keys/YOUR_API_KEY/stats \
  -H "Authorization: Bearer YOUR_API_KEY"

Response includes:

• Total requests (local vs. cloud)
• Cost savings
• Cache hit rate
• Rate limit status

Deployment

Railway

railway up

Fly.io

fly deploy

Self-Hosted

Use the included Dockerfile and docker-compose.yml for custom deployments.

Tech Stack

• Framework: FastAPI (Python 3.11+)
• Local Models: faster-whisper, llama-cpp-python
• Cloud APIs: OpenAI, Anthropic
• Database: PostgreSQL + SQLAlchemy + Alembic
• Cache: Redis
• Deployment: Docker, Railway, Fly.io

Configuration

Key environment variables:

Variable	Description	Default
LOCAL_MODELS_ENABLED	Enable local model processing	true
MAX_LOCAL_REQUESTS	Concurrent local requests before fallback	5
CACHE_TTL	Cache expiration in seconds	3600
RATE_LIMIT_WINDOW	Rate limit window in seconds	60

See .env.example for complete configuration options.

License

MIT License - see LICENSE file for details.

---

Built for developers who value privacy without sacrificing reliability.