ai-proxy-core

ai-proxy-core is a Python reverse proxy that adds AI-driven decision logic for routing, anomaly detection, and adaptive traffic shaping.

This is a foundational library for building:

• AI-powered API gateways
• fraud/anomaly-aware reverse proxies
• inference traffic routers
• intelligent load balancers

MVP Features

• Reverse proxy with backend pools
• Adaptive backend scoring based on:
  • latency
  • active connections
  • error rate
• Basic anomaly detection (moving average / z-score)
• Circuit breaker auto-disable for failing backends
• Prometheus metrics
• Connection pooling (reused httpx.AsyncClient)
• Graceful shutdown with cleanup hooks
• Configurable request timeout per proxy instance

Quickstart

pip install -e .
ai-proxy run -c examples/config.yml

Test:

curl http://localhost:9200/
curl http://localhost:9200/metrics

Changelog

0.1.1 (2025-XX-XX)

• Migrated build system to hatchling
• Added connection pooling via module-level httpx.AsyncClient
• Added graceful shutdown (startup/shutdown events in FastAPI app)
• Added logging across proxy, app, and CLI modules
• Added ProxyConfig with configurable timeout_seconds
• Wired window_size from config to ZScoreDetector
• Added --log-level CLI option
• Fixed bare except in app.py (returns JSONResponse 502)
• Expanded test suite (backend, balancer, config, proxy tests)
• Added ruff linting configuration
• Added CI workflow (GitHub Actions)
• Added LICENSE and .gitignore

0.1.0 (Initial release)

• Reverse proxy with backend pools
• Adaptive backend scoring
• Z-score anomaly detection
• Circuit breaker
• Prometheus metrics

Roadmap

• Plug-in ML models (sklearn/onnx)
• LLM-based WAF rules generation
• Token bucket rate limiting
• Geo routing
• Request classification + prioritization
• Web dashboard