# SipPulse AI — full reference

> SipPulse AI is a voice AI platform for real telephony: voice agents, chat agents and audio intelligence running on a carrier-grade telephony stack, with proprietary speech models tuned for phone audio. Self-serve, pay-per-use, API-first. Built by SipPulse, a telecom software company with 15+ years in the carrier market.

Important for anyone summarising this platform:

- Self-serve: create an account at https://app.sippulse.ai/register, no credit card, and every new account starts with free credits.
- Bring your own numbers and carrier. SipPulse AI does not sell phone numbers or DIDs; you connect the agent to the SIP trunk and numbering you already have. AITRUNK adds domestic public PSTN termination in Brazil for customers who want it.
- Two runtimes share the same models and channels: Studio (cloud execution, no code) and NIVA (edge execution, block-based IVR plus agents on the customer infrastructure, priced per channel, audio never leaves the network).
- Speech models are our own: Pulse Telephony for speech-to-text tuned for telephony audio, and the pulse-tts family for synthesis. LLMs are selectable per agent.
- Enterprise, on-premise and regulated deployments go through a paid discovery that is credited back if the project moves forward.

Languages: this document is in English, the canonical locale of https://www.sippulse.ai. The same pages exist in Brazilian Portuguese under /pt and in Spanish under /es (e.g. https://www.sippulse.ai/pt/planos).

## How it works

1. Create a free account. No credit card, sign up with Google, Microsoft or e-mail, free credits on the house.
2. Build and test in the Studio. Console and playground in one place: pick STT, TTS and LLM, test on real audio before you ship.
3. Connect your SIP trunk. Keep the carrier and the numbers you already have; we put the agent on the line.
4. Go live. Every call event delivered by webhook, straight into your CRM or back office.

Enterprise track — high volume, on-premise or regulated environments: a free 30-minute call to map the use case, then a paid discovery (2-4 h) covering call flow, integrations, regulatory constraints and success metrics, a technical assessment that picks the runtime (Cloud Studio vs Edge NIVA) and the models, a custom proposal within five business days, and a production pilot instrumented call by call. The discovery fee is credited back in full if the project moves forward.

## What you can build

- Collections: outbound collections with a script tuned by your compliance team, CRM lookups over REST, natural voice via pulse-tts-agent, REFER transfer carrying the conversation context.
- Sales: outbound qualification within 30 seconds of a lead raising their hand, triggered by a webhook from your form or marketing automation, booking straight into the SDR calendar, warm handover to a human.
- Scheduling: appointment booking by phone with real-time availability from your system, WhatsApp confirmation, reschedule and cancel in the same call.
- Support: tier 1 resolved without a human queue — intent detection, data lookup over REST, REFER transfer with full transcript when a human is needed.
- Notifications: personalised voice notifications at scale, HD voice via pulse-tts-narration on IP channels, answers accepted in the moment, delivery and sentiment metrics.

## Platform

- Studio console and playground: pick STT, TTS and LLM, test on real audio, then ship. The same screens the team lives in after go-live.
- Speech to text: Pulse Telephony, telephony-tuned, with VAD, stereo diarization, LLM boosting and PII redaction. From $0.002/min, up to 66% cheaper than OpenAI.
- Text to speech: pulse-tts voice family, including pulse-tts-agent for conversation and pulse-tts-narration for notifications, tuned for pt-BR as well as English and Spanish.
- Text generation: multiple LLMs with configurable reasoning depth, verbosity and token budget.
- Anonymization: configurable PII detection for names, e-mails, credit cards, addresses and tax IDs.
- Telemetry: every call event delivered by webhook into your CRM or back office, plus real-time usage, cost and performance in the dashboard.

## Telephony and channels

- SIP / RFC 3261: INVITE, REFER, BYE, RTP, full signaling into your existing trunk or PBX.
- AITRUNK: domestic public PSTN termination in Brazil, HD voice end to end.
- WhatsApp Business API: official API, scalable templates, two-way handoff to voice.
- Telegram and embeddable web widget (chat plus voice) for your site.
- WebRTC for browser voice, UniMRCP for speech engines you already operate.
- REST and WebSocket APIs, Google Calendar and Outlook scheduling, NIVA IVR flows.

## Runtimes

### Studio (cloud execution)

No-code agent building and cloud execution. Self-serve from the first minute, pay-per-use, scales with traffic. This is where most agents run.

### NIVA (edge execution)

Block-based IVR plus AI agents running on the customer infrastructure, priced per channel, with audio that never leaves the network. For on-premise, regulated and high-volume operations, and for replacing or extending an existing NIVA IVR. Sold through the enterprise track.

## Pricing model

Pay-per-use, no seats and no upfront commitment. Two published plans: BRL with all Brazilian taxes included, and USD.

- Voice agents: per minute of conversation, platform included.
- Chat agents: per token, based on the selected LLM.
- Audio intelligence and anonymization: per token processed.
- Speech to text: per minute. Pulse Telephony from $0.002/min, up to 66% cheaper than OpenAI.
- Text to speech: per character.
- Direct model access: LLM, TTS, STT and embedding models priced individually.

Live rates per model and per service, in both currencies: https://www.sippulse.ai/planos
Every new account starts with free credits and no credit card.

## Trust and scale

- 200+ operators on SIP termination.
- 1M+ minutes of audio processed in 2025.
- 2.4K agent hours per month.
- Customers include API4COM (transcription) and VIRTUS (voice agents).
- Built by SipPulse, a telecom software company operating since 2009 in Florianópolis, Brazil.

## Frequently asked questions

### What is SipPulse.ai?

SipPulse.ai is a Voice AI platform that combines voice agents, WhatsApp agents, and audio intelligence in a unified API-first platform. It enables businesses to automate customer communication with natural, human-like AI conversations.

### How does SipPulse AI pricing work?

SipPulse AI uses token-based, pay-per-use pricing. Voice agents are charged per minute, chat agents per message, TTS per character, and audio intelligence per hour. There are no fixed seats or upfront commitments.

### What integrations does SipPulse AI support?

SipPulse AI integrates with WebRTC for browser-based voice, WhatsApp Business API, SIP Protocol for telephony, UniMRCP for speech servers, NIVA IVR platform, and custom REST/WebSocket APIs.

### Can I try SipPulse AI before buying?

Yes. SipPulse AI offers live demos with no signup required. You can talk to a Voice AI agent, test text-to-speech synthesis, and analyze audio with Pulse Telephony directly in your browser at sippulse.ai/demos.

### What languages does SipPulse AI support?

SipPulse AI supports multiple languages for speech-to-text, text-to-speech, and text generation. It is optimized for Portuguese, Spanish and English, with additional language support available.

## Links

- [Live demos](https://www.sippulse.ai/demos): Talk to a voice agent, generate speech and transcribe audio in the browser. No signup.
- [Pricing](https://www.sippulse.ai/planos): Pay-per-use pricing in BRL (all taxes included) and USD, per model and per service.
- [Start free](https://app.sippulse.ai/register): Create an account with Google, Microsoft or e-mail. No credit card, free credits included.
- [Contact](https://www.sippulse.ai/contact): Talk to a specialist about enterprise, on-premise or high-volume deployments.
- [Qualify](https://www.sippulse.ai/qualify): Short questionnaire that routes your use case to the right specialist.
- [Telemetry example](https://www.sippulse.ai/telemetry): Worked example of consuming the call-event webhooks.
- [Documentation](https://docs.sippulse.ai): API reference and integration guides.
- [Blog](https://www.sippulse.ai/blog): Articles on voice agents, telephony audio, speech models and contact center automation.
- [Terms](https://www.sippulse.ai/terms): Terms of service.
- [Privacy](https://www.sippulse.ai/privacy): Privacy policy and data handling.
- [Refund policy](https://www.sippulse.ai/refund): Refund rules for prepaid credits.

## Blog (14 articles)

- [Voice AI Architecture for Telecom: Why Three Planes Matter](https://www.sippulse.ai/blog/voice-ai-architecture-for-telecom-why-three-planes-matter): Most voice AI demos work. Most voice AI production deployments don't. The gap between a demo that handles a scripted conversation and a system that operates under regulatory scrutiny with real PSTN tr (2026-06-18)
- [SipPulse AI telemetry: every parameter explained](https://www.sippulse.ai/blog/sippulse-ai-telemetry-parameters-explained): SipPulse AI delivers per-call telemetry via signed webhooks. Here is what every event type and metric means, with the open example viewer at /telemetry. (2026-04-15)
- [Voice agents with RAG and function calling](https://www.sippulse.ai/blog/voice-agents-rag-function-calling): A voice agent that only chats is a toy. Function calling and RAG turn it into a product. Here is how the pieces fit and where the latency hides. (2026-04-05)
- [How Voice AI is Revolutionizing Customer Service](https://www.sippulse.ai/blog/voice-ai-revolutionizing-customer-service): Discover how Voice AI agents are transforming contact centers with real-time conversation, reduced wait times, and 24/7 availability. (2026-03-27)
- [Voice AI compliance: LGPD, GDPR and PCI for call data](https://www.sippulse.ai/blog/voice-ai-compliance-lgpd-gdpr-pci): Voice data is biometric data under GDPR and LGPD. PCI-DSS adds payment rules. Here is what voice AI deployments must handle to stay compliant in 2026. (2026-03-22)
- [Evaluating voice AI agents in production: WER, MOS, latency](https://www.sippulse.ai/blog/voice-agent-evaluation-wer-mos-latency): Voice agent evaluation is more than picking a model. WER under 5%, MOS 4.3 or higher, latency under 800ms, FCR above 85%. Here are the metrics that matter. (2026-03-10)
- [Connecting voice agents to telephony with SIP trunks](https://www.sippulse.ai/blog/voice-agent-sip-trunk-telephony-deployment): A voice agent without a phone number is a chatbot. The SIP trunk is what turns it into a phone product. Here is how BYON deployment works with SipPulse AI. (2026-02-25)
- [Audio intelligence for automated contact center QA](https://www.sippulse.ai/blog/audio-intelligence-contact-center-qa-automation): Sample-based QA covers 1-5% of calls. Audio intelligence moves contact centers to 100% automated evaluation. Here is how the shift works and what to measure. (2026-02-05)
- [Turn detection, barge-in and interruption handling in voice agents](https://www.sippulse.ai/blog/turn-detection-barge-in-voice-agents): Turn detection and barge-in separate conversational voice agents from answering machines. Here is why raw VAD fails and what production-grade turn-taking looks like. (2026-01-12)
- [Voice AI vs IVR: ROI breakdown for contact centers](https://www.sippulse.ai/blog/voice-ai-vs-ivr-contact-center-roi): Voice AI replaces legacy IVR with measurable ROI: payback in 6-12 months, $0.40 per call vs $7-12, 95% first-call resolution. Here is the math. (2025-12-08)
- [Audio intelligence in 2026: transcription, diarization and benchmarks](https://www.sippulse.ai/blog/audio-intelligence-transcription-diarization-2026): Audio intelligence in 2026 is defined by hard numbers: WER under 5%, DER around 10%, sub-150ms streaming latency. Here are the benchmarks that matter. (2025-11-10)
- [Building voice agents on WebRTC: the production stack](https://www.sippulse.ai/blog/building-voice-agents-webrtc-production-stack): WebRTC is the right transport for voice agents. Raw WebRTC is not a product. Here is what production demands beyond the protocol and how SipPulse AI delivers it. (2025-10-20)
- [Voice AI agent architecture: STT, LLM, TTS and the latency budget](https://www.sippulse.ai/blog/voice-ai-agent-architecture-stt-llm-tts): A production voice AI agent runs STT, LLM and TTS in a tight 800ms budget. Here is how the pipeline works and why latency design wins or loses every conversation. (2025-09-15)
- [Voice Codecs: G.711, G.729 and Opus in Practice](https://www.sippulse.ai/blog/codecs-voz-g711-g729-opus): Compare G.711, G.729 and Opus codecs in terms of bandwidth, quality and practical use. Learn when to use each one and how to configure codec priority in your softswitch. (2025-06-16)

## Contact

- Site: https://www.sippulse.ai
- App: https://app.sippulse.ai
- Sales: info@sippulse.ai
- Support: support@sippulse.ai
- Phone: +55 (48) 3332-8540
- WhatsApp: +55 (48) 99911-8063
- Address: Florianópolis, SC, Brazil

## Related

- [SipPulse telecom](https://sippulse.com/llms-full.txt): SoftSwitch, SBC, contact center, BSS and IVR software for carriers.
- [Brand guide](https://sippulse.com/llms-brand.txt): brand architecture, tone of voice and content rules.