Sales deck · July 2026

Voice agents
that live inside
your product.

The code-first platform for voice, chat and WhatsApp AI agents. One SDK, five channels. Your code, your database, your rules.

SDKv0.3.2
ChannelsPhone · WhatsApp · WebRTC · Chat · SIP
LatencySub-second
Languages60+
01 — The state of the art

Voice platforms have it backwards.

Traditional platforms
  • 01You configure the agent in their dashboard.
  • 02You define tools as JSON schemas.
  • 03You expose public webhooks.
  • 04Your app bends to the platform.
Pinecall
  • 01The agent is your code.
  • 02Tools are local functions.
  • 03Zero new public surface.
  • 04The platform bends to your app.
02 — Division of labor

You own the what.
Pinecall runs the how.

Your code
  • Prompts and agent personality
  • Tool functions with access to your DB
  • Business logic and validation
  • Conversation history
  • When calls start and end
The voice server
  • Audio transport: WebRTC, Twilio, SIP
  • Speech-to-text: Deepgram, Gladia, AWS
  • Text-to-speech: ElevenLabs, Cartesia
  • Voice Activity Detection and turn detection
  • Mixing, barge-in and live streaming
03 — Architecture

One agent. All the channels.

Same prompt, same tools, same code. Adding a fourth channel is one line.

01 · Voice
Phone
Twilio managed or BYOC. Numbers in 100+ countries.
02 · Voice
WebRTC
In-browser widget. Peer-to-peer, sub-second.
03 · Text
WhatsApp
Meta Cloud API. Transcribes voice notes automatically.
04 · Text
Web chat
Widget or headless. Escalates to a call without losing the thread.
05 · Voice
SIP
Enterprise trunks. Plugs into your existing PBX.
04 — Live demos
Call it.

Three channels, three production agents running right now. Pull out your phone, scan the code, talk to the browser.

Demo 01 — Phone call

One number.
An agent that already knows who you are.

01
Dial {{ phoneNum }}. You'll talk to Pines, a weather agent running on OpenAI Realtime.
02
Ask it the weather in any city. It runs real tools against Open-Meteo while it talks.
03
Cut it off mid-sentence. Barge-in is native to the model, not a VAD trick.
04
Switch it to Spanish. Voice, STT and language hot-reload without hanging up.
Twilio · {{ phoneNum }} On call
P
Pines · Weather Agent
Pines
It'll be clear in Buenos Aires with a high of 18 degrees. Want the extended forecast?
get_weather({"city":"Buenos Aires"}) → clear · 18°
00 : 42
mic
alt
×
Demo 02 — Web chat

Chat that escalates
to a call without cutting the thread.

01
You start by typing. Same agent, same tools, same prompt.
02
You tap the mic button. The conversation escalates to WebRTC and the transcript travels with you.
03
You refresh the page. The thread persists server-side and picks up where you left off.
04
Voice notes are transcribed into editable text in the input, WhatsApp style.
Chat · florencia · agent Streaming
Hi, I'm Florencia. I can book you an appointment, tell you about treatments or put you through to reception. How can I help?
I need an appointment with Dr. Vega for next week
Great. I have Tuesday 10:30, Wednesday 15:00 and Friday 09:15. Which works best?
Wednesday at 15:00
Send
Demo 03 — WebRTC in the browser

Voice on your site.
Zero plumbing.

01
Three lines of HTML. A web component that runs in React, Vue, Astro or plain HTML.
02
The agent draws UI. Tool calls render cards, forms, maps and pickers live.
03
Identity sealed in the token. Real multi-tenancy without trusting the prompt.
04
Audio goes peer-to-peer. Your backend never proxies audio.
WebRTC · voice.pinecall.io Connected
Hi, I'm Mara. Tell me what you need and we'll sort it out right now.
llm.tool_result · get_forecast
Buenos Aires · clear · 18°
7-day forecast
Talk
Mute
End
05 — Sectors
One engine.
Every vertical.

The same SDK serves a clinic, a bank, an ecommerce and an office tower. Change the prompt and the tools. Nothing else.

Verticals · use cases

Six industries. The same engine.

01 · Healthcare
Scheduling and triage
24/7 phone reception that books, cancels and reschedules against the real calendar. Escalates emergencies to a human on call.
Fillers · Blossom
02 · Retail
Post-sale and tracking
Order tracking, returns and stock questions over WhatsApp and phone. Queries the ERP live.
Ecommerce · marketplaces
03 · Real estate
Front desk and visitors
AI intercom that identifies visitors, opens doors, notifies the host and escalates to security.
Towers · corporate buildings
04 · Financial
Conversational CRM
Lead qualification, soft collections and deal confirmation with identity sealed in the token.
LumiCRM · Digiclarity
05 · Industrial
Field service
Scheduled outbound callbacks, satisfaction surveys and technician dispatch driven by customer history.
Naffco · TPX
06 · Hospitality
Bookings and concierge
Table bookings, pre-check-in and service questions in the guest's language, with hot reload per call.
Restaurants · hotels
Case 01 · LumiCRM

A CRM that
answers the phone.

01Bookings and reschedulingVoice · WhatsApp · Web · IG
02Filling cancelled slotsOutbound
03Collections on overdue accountsOutbound
04A living profile of every clientMemory
05Human takeover with contextPause / Resume
LUMICRM · AUTONOMOUS OPERATOR FOR SPAS AND SALONS
One Lumi. Four channels. Same memory.
Lumi answers in her own voice, replies on WhatsApp, chats on the site and handles DMs — against the same calendar and the same client profile. A single shared agent serves every location: each salon's identity travels sealed inside the token, and tenant isolation lives in code, not in the prompt.
Resolution96% of calls with no human Scale62 locations running on Lumi Impact+22% occupancy in 90 days StackSDK + WebRTC + WhatsApp + sealed tokens
Case 02 · Axion

Your quant platform,
on the phone.

01Alerts that call youNative ring
02Position status by voiceTool call
03Quick questions over chatChat
04Live transcript, word by wordDataChannel
05Controls from the lock screenCallKit
AXION · QUANT MONITORING WITH PINECALL AGENTS
When the market moves, the agent calls you.
Axion tracks a quant platform in real time. When a signal crosses the threshold, the agent fires a native call to the iPhone — system ring, iOS call screen — and reports the position by voice, with the figures coming live from the platform. The rest of the day, the same agent answers over chat inside the app.
PlatformNative iOS · @pinecall/ionic ExperienceCallKit · WhatsApp-style call DataTools against the quant platform, live VerifiedEnd-to-end on a real iPhone
Mobile SDKs

Native calls
inside your app.

01
A real system ring. CallKit on iOS, Telecom on Android — the agent's call looks and sounds like a real call.
02
Headless by design. The SDK ships no UI: the call screen is 100% yours, with your brand.
03
Live transcript over DataChannel — you watch it write itself word by word while the agent speaks.
04
Ionic and React Native. The API key never touches the device: your backend mints 60-second tokens.
screenshot Ionic app · CallKit
screenshot React Native app · Axion
The technical difference
The agent lives inside your system. It uses your database, your internal APIs, your hardware. No public webhooks. No new surface.
Division of responsibilities · Pinecall SDK technical dossier
What an agent looks like in Pinecall

A full agent
in 30 lines.

agent.ts 
import { Pinecall, tool } from "@pinecall/sdk";
import { db } from "./db.js";

const pc = new Pinecall();

const lookupOrder = tool({
  name: "lookupOrder",
  description: "Look up an order by the customer's phone number.",
  execute: async ({ phone }) => await db.orders.findOne({ phone }),
});

export const agent = pc.agent("support", {
  prompt: "You're Acme's support agent. Be brief and human.",
  llm: "openai/gpt-5.3-chat-latest",
  voice: "elevenlabs/valentina",
  stt: "deepgram/flux",
  phoneNumber: "+13186330963",
  greeting: "Hi, how can I help?",
  tools: [lookupOrder],
});
// that's it. phone, chat, whatsapp and webrtc — same agent.
06 — Production confidence

Your agent goes through QA
before it answers.

You write specs in YAML. An LLM judge talks to your agent over real voice or over text and reports pass / fail in a CI pipeline. You can record the call, listen to it and audit it.

  • 01Real voice mode: two agents bridged with STT, TTS and barge-in.
  • 02Text mode: 10× cheaper, ideal for fast regression.
  • 03Exit code for CI, JSON output for dashboards.
$ pinecall test agent/specs --voice   ━━━ date-handling.spec.yaml ━━━ Turn 1: "Hi, what day is it today?"   Bot: Today is Wednesday, July 29. Turn 2: "Book it for tomorrow."   checkAvailability({"date":"2026-07-30"}) ✓ pass 4.3s · 2 turns   ━━━ summary ━━━ passed 3 / 4 avg 7.9 / 10
07 — Why Pinecall

What no other platform gives you.

Hot reload
Change voice, language, model and prompt during a call.
Live A/B testing. Automatic multi-language. No redeploy.
Skills
An agent that loads capabilities on demand.
Bookings, billing, tech support — each skill brings its own tools and its own KB. Less hallucination, less latency.
Hybrid RAG
Documents + BM25 + embeddings, with citations.
Upload Markdown and the agent answers with the sources visible to the user.
Human takeover
Pause the AI. Take control. Hand it back with context.
Dashboard ready. WhatsApp today, voice on the roadmap.
Real multi-tenancy
Identity sealed inside the token.
Isolation between customers lives in your code, not in the prompt.
Live listening
One link. Any supervisor. The call in real time.
Optional recording. Nothing to install.
08 — Product status

In production, today.

60+
languages supported across STT and TTS.
30
interchangeable STT, TTS and LLM providers.
5
channels unified per agent.
80KB
SDK footprint. One runtime dependency.
cloudacio
digiclarity
Fillers
Blossom
Naffco
TPX
Axion
LumiCRM
09 — Commercial model

Pay-as-you-go credits.
No surprises.

Free trial
To try the SDK in 5 minutes.
No managed numbers. No knowledge bases. No outbound. Ideal for a local demo.
Starter
Up to 2 production numbers.
Knowledge bases, WhatsApp, outbound campaigns. A plan to go live with.
Pro / Enterprise
Unlimited numbers, BYOC Twilio, SLA.
Consolidated billing, BYOK provider management, dedicated technical support.

Managed: usage drawn from your credits. BYOK: your provider bills you directly, zero Pinecall markup. You choose per model, per provider and per agent.

Next steps

Let's talk.
Literally.

Dial the number, scan the widget, or send us a message. In 48 hours we'll have a prototype agent running against your database, inside your own process.

{{ phoneNum }}
pinecall.io
10 — Knowledge bases

You upload a PDF.
The agent knows it.

One knowledge base per agent or per skill. Upload PDFs, Word, Markdown or HTML and Pinecall chunks, embeds and indexes them. The agent searches and cites — mid-call, with no perceptible latency.

  • 01Postgres + pgvector. One SQL database: documents, chunks and vectors in the same transaction. No separate search service, no pipeline to maintain.
  • 02Perplexity Context embeddings. A state-of-the-art retrieval model — accuracy well above generic embeddings, especially on technical, legal and product jargon.
  • 03Hybrid BM25 + vector. The literal (a SKU, a policy number, a code) is found by the lexical side; the intent, by the vector. Reranking over both.
  • 04Citations, always. The answer carries the snippet and the source document. If it isn't in the KB, the agent says so — it doesn't improvise.
$ pinecall kb push ./docs --agent support   parsing condiciones-2026.pdf · 48 pp chunking 312 chunks · overlap 80 embedding perplexity/context ✓ indexing pgvector + bm25 ✓ 1.8s   ━━━ live query ━━━ user: "does it cover hail damage?"   kb.search({"k":4,"hybrid":true}) · 41ms   0.94 condiciones-2026.pdf §7.2   0.81 anexo-coberturas.md §3   bot: Yes, with a 5% deductible.   source: condiciones-2026.pdf §7.2
Demo settings
Press S at any time to open this panel. The number is saved for your next meeting.
{{ pageCur }} / {{ pageTot }} — {{ pageLabel }}
Intro Demos Verticals Product Contact