Observed Signal · Apr 25, 2026 · Technical Release · Source: DEV Community · Impact: 3/5 · Sentiment: Positive
Open-source AI for Real PSTN Phone Calls
CallPilot is an open-source FastAPI server that makes outbound PSTN phone calls using realtime audio LLMs. The system bridges Twilio's WebSocket media stream with AI voice providers (OpenAI Realtime and Google Gemini Live) via a provider abstraction, uses per-client RAG (ChromaDB) to ground responses, and implements interruption handling, voicemail detection (Twilio AMD), and call recording. The repo (github.com/kennedyraju55/callpilot) includes the bidirectional WebSocket audio bridge, audio transcoding logic for Gemini, and a mid-call RAG re-querying prototype. The author reports ~ $0.07–$0.23 total cost per ~2-minute call depending on provider and notes the project is MIT-licensed. Publication date: 2026-04-25.
Practical, open-source architecture and cost benchmarks for realtime voice agents; useful reference for developers building conversational/agentic voice automation but not a major platform policy or market shift.
Track Twilio Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- CallPilot is an open-source FastAPI server hosted at github.com/kennedyraju55/callpilot.
- The system places outbound PSTN calls via Twilio and streams media to the server over a Twilio WebSocket.
- CallPilot supports a provider abstraction for realtime audio LLMs and has working integrations for OpenAI Realtime and Gemini Live.
- Per-client RAG uses ChromaDB: documents are chunked, embedded, and the top-5 chunks are injected into the system prompt to reduce hallucinations.
- Estimated cost for a ~2-minute call: OpenAI path ~$0.13–$0.23 total; Gemini Live path ~$0.07–$0.11 total (Twilio voice ~ $0.03 + realtime LLM cost).
Connected Companies & Entities
5 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
OpenAI Launches GPT-Live-1 Voice Model in API
OpenAI has launched GPT-Live-1 in its API, a voice-native model designed for building voice-enabled applications and business workflows. The model handles listening and speaking simultaneously, reducing latency by eliminating the need for chained speech-to-text, reasoning, and text-to-speech components. It features improved interruption handling, background noise resilience, and the ability to delegate reasoning and tool calls to backend models like GPT-6 Astra or third-party models. The API includes telephony support for full-duplex voice agents. Pricing is set at $0.05 per minute for the voice layer, with backend models billed separately. Early evaluations show an 80% reduction in interruptions for language tutoring compared to turn-based systems.
Open-source AudioTrace Reveals AI Voice Agent Signals
A developer published AudioTrace, a small open-source library that analyzes voice-agent call recordings and emits a structured, typed CallReport. The tool splits audio analysis into two layers: deterministic signal-processing measurements (silence, pitch, speaking pace, latency) and learned-model estimates (transcript, speaker, sentiment, intent). AudioTrace runs locally for privacy, can be installed via pip, returns a Pydantic CallReport, and is designed to integrate with observability tooling (OpenTelemetry) and agent tracing systems (LangChain, LangSmith). The blog post is the first of a series that will detail signal-extraction approaches and how to wire voice signals into CI pipelines. The project repository is published on GitHub (github.com/dimastatz/audiotrace).
How to detect an AI-generated voice on live calls
New blog post published on Sep 21, 2026, exploring how AI voice detection works on live phone calls, from PSTN codecs and streaming detection to post-call audits and watermarking.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.