Skip to main content

​
Callvrse

Callvrse is a programmable voice AI media gateway and telephony platform. It connects the phone network — carriers like Telnyx and Twilio, or your own Asterisk/Kamailio PBX — to AI voice agents that answer, listen, and speak in real time over AudioSocket and WebSockets. The gateway handles the audio plumbing while your agents handle the conversation. It performs voice activity detection in under 50ms so agents can barge in mid-sentence, resamples audio between telephony and voice-AI formats on the fly, records both sides of every call, and streams live diarized transcripts to your dashboard.

​
Core capabilities

​
Carrier integration and media bridging

  • Inbound and outbound call routing across Telnyx, Twilio, and BYO Asterisk PBX
  • Least cost routing (LCR) that picks the cheapest or best-quality carrier at dial time
  • Bidirectional media bridging between AudioSocket (linear 16-bit PCM) and AI voice agents over full-duplex WebSockets
  • STIR/SHAKEN caller identity signing and inbound attestation verification (levels A, B, C)
  • WebRTC browser signaling gateway for agent consoles

​
Audio and voice AI

  • In-gateway resampler converting between 8 kHz (telephony), 16 kHz (wideband), 24 kHz (TTS such as Cartesia and ElevenLabs), and 48 kHz (Opus WebRTC)
  • In-gateway Silero VAD with adaptive noise-floor calibration and debouncing: on speech detection the gateway cuts agent playback and emits barge-in events
  • Native handling of linear PCM (8k/16k/24k/48k), G.711 μ-law (PCMU), G.711 A-law (PCMA), and G.722

​
Recording and transcription

  • In-memory dual-stream recording of caller (left) and agent (right) audio into a stereo RIFF WAV
  • In-process AWS SigV4 client uploading recordings to S3, MinIO, or Cloudflare R2 without blocking call teardown
  • Public recording URLs attached to call detail records (CDRs) in PostgreSQL or SQLite
  • Deterministic diarization: caller and agent audio live on separate buffers, so speaker identification has no cross-talk ambiguity
  • Multi-provider STT (Gemini, Deepgram, Cartesia, Whisper) with WebVTT output and live SSE utterance events

​
Reliability

  • Redis-backed distributed call presence across gateway and API replicas
  • Distributed token-bucket rate limiting via atomic Lua scripts
  • Kamailio auto-failover to a secondary media switch on unavailable or 5xx destinations
  • Graceful call draining on rolling updates, keeping in-progress calls bridged up to a configurable timeout

​
Supported agent runtimes

Configure your agents through the dashboard or the agents API: The Speech Engine mode is a half cascade: ElevenLabs owns speech-to-text and text-to-speech, while your chosen LLM composes the replies from a Pipecat pipeline running in the agent service.

​
Start here

  • Browse the API Reference tab for every endpoint, generated from the OpenAPI spec (openapi.json).
  • Running the gateway locally: see local development in the README.
  • Deploying to production: Docker Swarm stack in deployments/stack.yml.