SIPHON

Building a calling agent? Stop renting your infrastructure. If you've tried building a calling agent recently, you know the struggle. You either pay platforms like Vapi/Retell ~$0.08/min (a huge markup just to wrap OpenAI), or you waste weeks stitching together SIP trunks and WebSockets before writing a single line of logic. Meet Siphon. We built the third option: Open-source infrastructure that gives you the ease of a managed service with the economics of a DIY solution. Why developers are switching to Siphon: 💸 Kill the Wrapper Tax: Bring your own keys (OpenAI, Deepgram, Twilio). We don't charge a "per-minute" markup. Your costs drop by ~90% (from ~$0.08 to ~$0.01/min). ⚡ Sub-500ms Latency: Built on LiveKit (WebRTC). The conversation feels instant and human, with zero "awkward pauses." 🐍 Python Native: No drag-and-drop builders. Write your agent logic in pure Python. Manage state, context, and function calling with the code you already know. 🛠 Production Ready: Handles interruptions, turn-taking, and scaling automatically. See it in action: You can deploy a receptionist in literally 10 lines of Python: from siphon.agent import Agent from siphon.plugins import openai, deepgram agent = Agent( name="Receptionist", llm=openai.LLM(), stt=deepgram.STT(), instructions="You are a helpful scheduler..." ) if __name__ == "__main__": agent.start() We are open-source and looking for feedback from the PeerPush community. Check out the repo and let us know what you think! https://github.com/blackdwarftech/siphon