# VSIP

> VSIP is a real-time voice-events API for AI voice agents: semantic end-of-turn detection, speaker identity locked to an enrolled user, response gating (`speaker_verification.should_respond`) and identity-gated barge-in, over one WebSocket (`wss://api.vsip.online/v1/stream`). It also offers a Speaker Profile REST API (enroll/verify/identify voiceprints) and a Batch Analytics REST API (speaker timelines for recorded audio). VSIP does not do speech-to-text, LLM or text-to-speech — it decides when and whether those run.

To integrate VSIP into a project, read the complete integration guide first; it is self-contained and includes the rules that integrations most often get wrong.

## Integration

- [AI integration guide (full text)](https://vsip.online/llms-full.txt): protocol, events, control messages, MUST rules, copy-paste Python/TypeScript loops, framework adapters, REST APIs, close codes, test checklist
- [OpenAPI spec](https://vsip.online/openapi.json): REST endpoints — `POST /v1/stream/token` (short-lived browser tokens), `/v1/profiles/*`, `/v1/analyze/*`
- [Example voice agent](https://vsip.online/examples/voice-agent/README.md): runnable Python agent — VSIP + OpenAI STT/LLM/TTS ([agent.py](https://vsip.online/examples/voice-agent/agent.py), [requirements.txt](https://vsip.online/examples/voice-agent/requirements.txt), [env.example](https://vsip.online/examples/voice-agent/env.example))
- [Human-readable version of the guide](https://vsip.online/docs/ai)

## SDKs

- Python: `pip install vsip-sdk` (import `vsip`) — `VSIPClient`, `VoiceSession`, `ProfilesClient`, `AnalyzeClient`, `create_stream_token`; adapters for Pipecat, LiveKit Agents, Twilio
- JavaScript/TypeScript: `npm install vsip-sdk` — `VSIPClient` (`apiKey` server-side or `getToken` in browsers), `VoiceSession`, `createStreamToken`

## Optional

- [Human documentation](https://vsip.online/docs)
- [Live demo](https://vsip.online/voice-lab)
- [Pricing](https://vsip.online/pricing)