IntelliCall AI

⚠ Important — external service costs are NOT included in this purchase. IntelliCall AI is a self-hosted application that connects to third-party services such as Twilio, ElevenLabs, Deepgram, Sarvam AI, Google Gemini, LLM providers via OpenRouter, and payment gateways. Those providers bill you directly — you bring your own API keys and pay them separately. Full details in the Important notice section below.

Leads go cold in minutes. Your AI agent calls them in seconds.

A new lead fills in your form at 11pm. Tomorrow at 10am someone calls back, and by then they have booked with whoever answered first. IntelliCall AI closes that gap: voice agents answer your phone lines, call your lead lists, qualify each prospect and book the appointment while the caller is still on the line.

It runs on your own server, with your own keys, on Twilio or any SIP carrier you choose. You can talk to an agent on the live demo right now.

→ Talk to the live demo

Two ways to make money with one purchase

• Use it in your business. Every inbound call answered, every lead called, every booking on your calendar, without adding callers to the payroll.
• Sell it as your own SaaS. Plans, limits, trials, four payment gateways and full white-label branding are already built. Put your name on it, charge clients monthly and keep the revenue. Reselling access needs the Extended License.

Who buys IntelliCall AI

• SaaS founders who want a finished, sellable product instead of six months wiring telephony, speech and billing together
• Agencies adding AI calling to their retainers under their own brand, without a platform taking a cut of every minute
• Sales and growth teams that need speed-to-lead and booked meetings without a call centre
• Operators who want their own server, their own keys and their own carrier, with costs that follow provider rates

How a call works, step by step

1. The trigger. A prospect calls one of your numbers, a scheduled campaign reaches their name on the list, or your web form fires the instant-dial API.
2. The greeting. The agent answers with your opening line, in your chosen voice and language, using the lead's details from merge fields.
3. The conversation. Real-time speech recognition and the agent's model keep it on script. The caller can interrupt at any point and the agent stops to listen.
4. The answers. Questions about prices, hours or policies are answered from your knowledge base.
5. The outcome. The agent books an open slot, sends the caller to a person on your team, or ends the call politely when the conversation is over.
6. The follow-through. The booking lands in Google Calendar, a WhatsApp reminder goes out before the meeting, and the lead syncs to HubSpot, Slack, n8n or your webhook.
7. The review. Each call comes back with a 0–100 score, a summary, a transcript and a recording.

Try it before you buy it

The live demo is a full deployment of the current version. Open it, pick an agent and talk to it from your browser.

URL: https://intellicallai.wixzel.com/

Install it from your browser

Buyers get the Web VPS installer: point it at a fresh server and it installs IntelliCall AI for you. Updates use the same installer.

Everything in the download

Grouped so you can jump to what matters to you. Everything listed ships in version 11.9.

Voice agents and voice engines

• Unlimited agents for inbound and outbound calls, each with its own name, system prompt and opening line
• Five voice engines, chosen per agent:
  • Classic: Deepgram speech recognition, any model on OpenRouter, ElevenLabs voices
  • Deepgram Voice Agent: one connection for listening, thinking and speaking, with server-side turn-taking and Deepgram-managed LLM (no OpenRouter key). Speaks with your ElevenLabs voice or Deepgram's own Aura voices
  • Gemini Live: Google's gemini-3.8-live model hears, thinks and speaks in one step, on a single Gemini key (SIP numbers)
  • Sarvam AI: Saaras speech recognition, Sarvam-105B and Bulbul voices, built for Indian languages, on a single Sarvam key
  • Twilio Standard: basic Twilio voices with nothing extra to configure
• Voices: your own ElevenLabs voices including clones, native ElevenLabs Voice Library voices for the agent's language (added to your account when you pick one), Deepgram Aura voices per language, 37 Sarvam voices and 30 Gemini voices, with in-editor previews
• Languages: 16 on the classic engines, 11 for India on Sarvam (Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, English), and 97 on Gemini Live, which also follows the caller's language automatically
• Natural interruptions: the caller can cut in and the agent stops to listen, while coughs, line noise and a quick "okay" do not cut it off
• Ends calls properly: hangs up when the conversation is over, ignores line noise during the goodbye, ends calls after five silent minutes, and caps call length inbound and outbound

Voice quality and conversation controls

• Presets per agent: Fast, Balanced, Natural or Expressive, plus Default, which keeps existing agents exactly as they were
• Custom mode with ElevenLabs stability, similarity, style, speaker boost, speed, model and latency profile
• Turn-taking controls: end-of-turn timing, silence tolerance and interruption sensitivity
• Per-call voice diagnostics in the call log: response latency, interruptions, false voice-detection events and audio buffer stats

Browser testing

• Talk to any agent from the dashboard with your microphone, with a live transcript beside the conversation
• Uses the agent's real engine, voice, language, knowledge base and booking, so what you hear is what callers get
• Needs no phone number, isn't logged as a call, doesn't count against plan limits, and stops itself after five minutes

Telephony: Twilio and SIP

• Start on Twilio, or connect any VoIP/SIP carrier through the bundled Asterisk integration: host, port, transport, auth, caller ID, codecs and dial prefixes
• As many numbers as you need, each routed to an inbound agent, with a fallback so no caller reaches a dead line
• Human Transfer (SIP): the agent hands the caller to reception, sales or a named colleague, and takes the call back with full context if nobody answers. Ring times, attempt limits and every outcome are logged
• SIP health checks in plain English, a live SIP event stream and per-trunk test calls, so most problems are fixed without SSH
• Scanners probing unknown numbers are detected and banned automatically

Campaigns and outbound calling

• Bulk campaigns from CSV or Google Sheets, with scheduling, start and stop, live status and duplication
• Instant dial: POST /api/calls/dial calls one number with a chosen agent within seconds, no campaign needed
• Campaign pages show the agent's real engine and voice, and plan limits are checked on every dial

Appointments

• Agents book open slots during the call and follow the agent's language
• Edit, reschedule or cancel from the Appointments page
• Google Calendar sync for every booking, change and cancellation
• WhatsApp reminders to the client from 15 minutes to 24 hours ahead, timezone-aware, re-armed on reschedule and never sent twice

Leads and call intelligence

• Lead workspace with tags, custom fields, bulk import, bulk tagging and bulk delete
• Custom fields become merge tags such as {{appointment_date}} in the greeting and system prompt
• Every call scored 0–100 with budget, timeline and next steps, plus an AI summary, a full transcript and a recording
• One dashboard for calls, leads, campaigns and appointments

Knowledge base and templates

• Knowledge bases of business info, FAQs and context, built from your website in one step or written by hand
• 37 ready-made agents in 10 categories (support, scheduling, hospitality, sales, outreach, finance, healthcare, real estate, e-commerce, fitness), including 21 non-English versions

Integrations and API

• n8n with 3 ready-made workflows: form to instant call, call log to Google Sheets, qualified lead to CRM
• WhatsApp (Meta Cloud API) alerts, Slack notifications, HubSpot contact sync, Google Calendar and Google Sheets
• Webhooks with HMAC-SHA256 signed payloads and per-event toggles
• Personal API keys and a built-in reference for 12 endpoints with copy-ready curl examples
• Each user connects their own accounts, isolated per tenant

Run it as a SaaS

• Plans with limits on agents, campaigns, leads and monthly calls, limited or unlimited, enforced at call time
• Configurable free trials and full purchase history
• Stripe, PayPal, Razorpay and Dodo Payments in 7 currencies (USD, EUR, GBP, INR, AUD, ZAR, CAD)
• Each customer brings their own provider keys, so their usage is billed to them
• White-label: app name, brand colour across the whole site, logos and favicon
• Admin panel for users, plans, revenue, payment keys, integration credentials, legal pages and support tickets, with an update checker
• Marketing site included: landing page with pricing, FAQ, testimonials and integrations, plus About, Contact, Privacy and Terms pages

Dashboard, access and security

• Dashboard in English, Spanish, French, German and Arabic (full right-to-left), chosen per user
• Light and dark mode, per-user timezone and 12/24-hour time
• Email and password or Google sign-in; user, admin and superadmin roles
• Personal API keys stored hashed and SIP trunk passwords encrypted at rest

Own it instead of renting it

Hosted voice-AI platforms charge a fee on every minute, on top of what the speech and telephony providers already cost, and their name sits on your client's calls. With IntelliCall AI you pay the providers directly at their own rates, the platform carries your brand, it runs on your server, and you get the full source code. Pick a cheaper engine or carrier per agent whenever the numbers call for it.

Built for SaaS founders and agencies

• Launch it as your own hosted product under your brand
• Package tiers, enforce limits and grow monthly recurring revenue
• Deliver AI calling to clients as a managed service
• Support your customers through the built-in ticket system
• Extended License: required to sell access as a service under your brand, per Envato terms

Important notice (external service costs)

IntelliCall AI is a self-hosted application that integrates with third-party services such as Twilio, SIP carriers, ElevenLabs, Deepgram, Sarvam AI, Google Gemini, LLM providers via OpenRouter, and payment gateways. The purchase price does NOT include any credits, subscriptions or usage costs for these services. You create your own accounts and pay each provider directly, based on your usage. Pricing varies by provider.

Tech stack

• Backend: Node.js 20 and Express 5, with WebSocket media streaming
• Frontend: Next.js 16, React 19 and Tailwind CSS 4
• Database: MongoDB with Mongoose
• Telephony: Twilio Media Streams, and Asterisk (ARI) for SIP trunks
• Authentication: Passport.js with Google OAuth 2.0, email and password, and JWT

What's in the download

1. Complete backend source code (Node.js / Express)
2. Complete frontend source code (Next.js / React)
3. Interactive QuickStart documentation with visual guides
4. Installer and setup scripts
5. 3 n8n workflow templates
6. Environment templates (.env.example)

Frequently asked questions

Are API and call costs included?

No. You bring your own keys for Twilio or your SIP carrier, the speech providers and the LLM, and each provider bills you directly. Sarvam AI and Gemini Live agents each need only one key.

Can I hear how it sounds first?

Yes. The live demo runs the current version, and you can talk to an agent from your browser.

Which languages does it speak?

97 on Gemini Live, 16 on the classic engines and 11 for India on Sarvam. Voice lists are filtered to the agent's language.

Do I need Twilio?

No. You can connect any SIP carrier through the bundled Asterisk integration instead, or use both.

How hard is installation?

The Web VPS installer sets it up on a fresh server from your browser. The QuickStart guide and video tutorials cover manual setup, and paid installation is available.

Can I sell it to my own clients?

Yes, with the Extended License. Plans, payments and white-label branding are built in.

More answers in the FAQ section of the support page.

Every lead called in seconds, every call answered, on a platform you own. Open the live demo and talk to it.

CHANGELOGS

Version 12.0 | 8 November 2026

- NEW: A new design across the dashboard, sign-in and the public site, in light and dark mode and all five languages
- IMPROVEMENT: The Primary Color in Admin Settings now colours the whole app, and text colours adjust to stay readable in both themes
- IMPROVEMENT: Voice engines and quality presets are cards, and long voice and language lists are searchable
- IMPROVEMENT: Delete, disconnect and other confirmations are proper dialogs instead of browser pop-ups
- IMPROVEMENT: Leads can be filtered by clicking tag chips, each showing how many leads it has
- NOTE: This update replaces the frontend's UI components. If you have customised frontend files, back them up and re-apply your changes after updating

Version 11.9.2 | 7 November 2026

- NEW: SIP trunks can send numbers without the + sign, for PBXs that reject a leading +
- IMPROVEMENT: Lead numbers are converted to international form before dialling, using the calling number's country; a number that can't be read is refused with the reason in Call Logs
- IMPROVEMENT: The SIP trunk, phone number, lead, CSV import and campaign forms point out mistakes before you save
- FIX: A SIP trunk pointing at this server itself is refused on save and test, instead of failing every call with "cause 1"
- FIX: Inbound calls to a number with no agent are forwarded to its fallback number over the number's own trunk, instead of being hung up
- FIX: Analyze Now says why analysis failed instead of "Failed to analyze transcript"
- NEW: Gemini Live and Sarvam AI calls can be analysed on their own engine's key when there is no OpenRouter key
- FIX: The AI Score column no longer shows "Pending" forever; it says why a call wasn't analysed
- SECURITY: Only a call's owner and super admins can analyse it

Version 11.9.1 | 4 November 2026

- NEW: Brazilian Real (R$) currency option in Admin → Settings → General

Version 11.9 | 18 October 2026

- FIX: Sarvam AI calls no longer stay connected after the goodbye when there is background noise or the caller says "thanks, bye"
- FIX: Background noise no longer resets Sarvam's silent-call timeout, so abandoned calls on noisy lines are ended
- FIX: The campaign details page showed "ElevenLabs AI / Rachel" for Sarvam, Gemini and Deepgram agents; it now shows the real engine and voice
- NEW: Deepgram Voice Agent can speak with Deepgram Aura voices, listed live per language (Italian included), with no ElevenLabs key needed
- NEW: The ElevenLabs voice picker lists native Voice Library voices for the agent's language and adds the one you pick to your account
- IMPROVEMENT: Campaign calls are protected against "An application error has occurred" caused by a trailing-slash or missing BASE_URL
- UPDATE: Gemini Live moves to the stable gemini-3.8-live model, ahead of the 17 November 2026 shutdown of the preview models
- UPDATE: Deepgram Voice Agent uses multilingual Flux for 9 more languages, Flux for every English accent, and Deepgram's current language fields
- FIX: UK, Australian and Indian English no longer go to nova-2-phonecall (US English only) on the classic engine
- UPDATE: ElevenLabs voices load from the /v2/voices API (the legacy endpoint fails past 500 voices); Turbo v2.5 is marked deprecated
- FIX: The Sarvam voice preview sends the language_code field Sarvam now requires
- FIX: Email login on a Google sign-in account no longer fails with "data and hash arguments required"; Settings can now set a password for those accounts
- IMPROVEMENT: The backend logs the MongoDB database it connects to and warns when MONGODB_URI has no database name
- FIX: Call Logs show why a call failed in its row, and Details opens for failed calls (it was disabled for calls without a transcript)
- FIX: Gemini Live calls record Google's own reason when a session is refused or dropped, and hang up instead of leaving the caller in silence
- FIX: Outbound SIP calls that Asterisk would not dial, or the carrier rejected, show the reason in Call Logs
- IMPROVEMENT: Saving a SIP trunk warns you if Asterisk did not load it
- IMPROVEMENT: SIP usernames are limited to letters, numbers and . _ - +, and the form names any other character (such as the @ of an email address) as you type
- FIX: npm run build no longer fails with "next/font/google queries have exactly one entry"; the fonts now ship with the app

Version 11.8 | 4 October 2026

- FIX: The Create Agent drawer could crash with INVALID_MESSAGE: MALFORMED_ARGUMENT because of the {{name}} first-message placeholder; it now shows as plain text in all five languages

Version 11.7 | 1 September 2026

- FIX: Settings could refuse to save with "webhooks.events.callTransferred is not allowed", which blocked every field on the page
- NEW: Google Calendar and Sheets credentials are set in Admin - Integrations - Google, with no file to edit and no restart
- IMPROVEMENT: The admin panel lists the exact authorized redirect URIs to register in Google Cloud Console
- FIX: A wrong Google Client ID is caught before the redirect, naming the fix instead of Google's Error 401 invalid_client page
- IMPROVEMENT: Notification events added in future releases can no longer break the Settings page

Version 11.6 | 29 August 2026

- NEW: HubSpot connects with a service key from your own account. No developer app or callback URL
- NEW: Service keys are verified on save, naming any missing scope instead of failing later at sync time
- IMPROVEMENT: Existing HubSpot OAuth connections keep working, with nothing to reconfigure
- FIX: OAuth callbacks now resolve as HTTPS behind a reverse proxy, clearing invalid redirect URL errors
- FIX: The dashboard now ships .env.example, so the documented setup step works on a fresh install
- NOTE: HubSpot ended new public app creation on 23 June 2026. Service keys are the supported path

Version 11.5 | 29 August 2026

- NEW: Human Transfer — your AI agent can now hand a live caller over to a real person, so a caller who says "let me speak to someone" is put straight through instead of hitting a dead end
- NEW: Set up as many authorised destinations per agent as you need — reception, sales, support, a specific colleague — each with its own name and phone number, configured right inside the agent editor
- NEW: Pick a default destination, used whenever the caller simply asks for "a person" without naming a department
- NEW: If the destination is busy, declines, or nobody picks up within the ring time you set, the caller is handed straight back to the same agent, mid-conversation, with everything it already knew about them — they never get dropped and never have to start again
- NEW: Per-destination and per-agent ring times, plus a cap on transfer attempts per call so an agent can't keep re-dialling someone who isn't there
- NEW: Every transfer is recorded on the call in Call Logs — which destination, the outcome, how long it rang, how long the caller spoke to the human, and the carrier's reason if it failed
- NEW: Two new notification events, "Caller Transferred" and "Transfer Failed", available on Webhooks, Slack, WhatsApp and n8n
- IMPROVEMENT: Works on all four voice engines — Deepgram Voice Agent, Gemini Live, ElevenLabs (Classic) and Sarvam AI
- IMPROVEMENT: Your agent is never told the destination phone numbers, only their names — so it cannot be talked into dialling a number you did not authorise, and the number is checked on the server before any call is placed
- IMPROVEMENT: IntelliCall keeps control of the call from start to finish, and the recording keeps running through the handover, so the conversation with your colleague is captured too
- NOTE: Human Transfer runs over your SIP trunk. Agents on a Twilio number, and the in-browser agent test, do not offer it — the option is simply not presented to those agents
- NOTE: Human Transfer is off by default, so every existing agent keeps behaving exactly as it did in v11.0 until you turn it on

Version 11.0 | 28 August 2026

- NEW: Test any agent live in your browser — click Test on the AI Agents page, talk to the agent through your microphone, and hear exactly how it sounds before it ever calls a customer
- NEW: The browser test uses no phone line at all: no Twilio call, no SIP trunk, no phone number, and no call minutes — agents with no number attached can be tested too
- NEW: Test sessions run the agent's real configuration — its engine, voice, language, voice quality and conversation settings, knowledge base and appointment booking — so what you hear is what a caller gets
- NEW: Live transcript alongside the conversation, so you can read what the agent heard and said while you talk to it
- NEW: Gemini Live agents can finally be heard without owning a SIP trunk, by testing them in the browser
- IMPROVEMENT: The agent always opens the conversation in a test — the microphone stays closed until it has started speaking, so background noise can't cut off its greeting
- IMPROVEMENT: Test sessions are never logged as calls and never count against your plan, and stop automatically after 5 minutes so a forgotten tab can't run up usage on your AI provider accounts
- NOTE: Agents using the basic "Standard (Twilio)" voice can't be tested in the browser — that voice is produced by Twilio during a real call — so the Test button explains this and stays disabled until the agent is switched to a streaming voice engine
- NEW: Gemini Live voice engine — one Google model hears, thinks and speaks on the call, so replies come back fast and sound genuinely conversational; selectable per agent on SIP numbers
- NEW: 97 spoken languages on the Gemini Live engine, including Malayalam and other languages no other engine covers, with an Auto-detect mode that simply follows whatever language the caller speaks
- NEW: 30 Gemini prebuilt voices to choose from per agent, from bright and upbeat to mature and warm
- NEW: Gemini API Key field in Settings — the engine is fully self-contained, so this one key is all it needs (no Deepgram, ElevenLabs or OpenRouter)
- NEW: Gemini Live agents support appointment booking, knowledge bases and the agent ending the call naturally, just like the other engines
- IMPROVEMENT: The engine picker now explains exactly why an engine is unavailable — a missing key or, for Gemini Live, a non-SIP outbound number — instead of hiding options
- IMPROVEMENT: Attempting to run Gemini Live over a Twilio number is blocked with a clear message when saving the agent, and the call log records a clear error if it ever happens

Version 10.9 | 28 August 2026

- NEW: Voice Quality & Conversation section in every AI agent — presets (Fast, Balanced, Natural, Expressive) make different agents sound completely different with the same ElevenLabs voice; the Default preset keeps existing agents behaving exactly as before
- NEW: Custom / Advanced mode exposing the raw ElevenLabs controls per agent — stability, similarity boost, style, speaker boost and speaking speed, plus the ElevenLabs model (Turbo v2.5, Flash v2.5 or Multilingual v2) and a latency profile that trades response speed against voice quality
- NEW: Turn-taking controls per agent — end-of-turn silence, silence tolerance and interruption sensitivity, so a patient receptionist and a snappy sales agent can each take turns the way their callers expect, with built-in protection against false interruptions from coughs and line noise
- IMPROVEMENT: More natural speech in every language — agents on a Voice Quality preset send complete clauses to the voice engine instead of cutting at every comma, which audibly improves intonation and fixes clipped short words, especially in Spanish and other Romance languages
- IMPROVEMENT: Replaced ElevenLabs' deprecated streaming-latency parameter with their current generation controls, selected automatically from each agent's latency profile
- NEW: Voice Quality panel in Call Log details — every call now records response latency (AI first token, first speech), interruptions, false voice-detection events, audio buffer diagnostics and speech chunks, so call quality issues can be diagnosed instead of guessed at
- NEW: Agents on the Deepgram Voice Agent engine can now choose their ElevenLabs model per agent

Version 10.8.2 | 27 August 2026

- NEW: Dial Prefix field for SIP trunks — wholesale carriers that route on a technical prefix (e.g. 999 in front of the number) can now be used for outbound calls; enter the prefix your carrier gave you and IntelliCall dials it exactly the way they expect
- IMPROVEMENT: The Dial Prefix field is empty by default, so every existing trunk keeps dialling exactly as before — only fill it in if your carrier documents a prefix
- IMPROVEMENT: The current version number shown in Admin → Settings now tracks the release it ships with

Version 10.8.1 | 27 August 2026

- FIX: On SIP calls, agent replies longer than about 10 seconds could have a chunk silently cut out of the middle, and playback could sound rushed or skip parts of sentences — buffered speech is no longer discarded, and playback now stays in step with the call for replies of any length
- FIX: Applied the same fix to every SIP voice engine — ElevenLabs (Classic) and Deepgram Voice Agent now use the corrected audio buffering that Sarvam agents received in v10.7
- IMPROVEMENT: Long replies are buffered up to 60 seconds instead of 10, and the limit is now configurable with SIP_RTP_MAX_QUEUE for unusually long agent responses
- IMPROVEMENT: Removed two unused installer scripts that were not part of the documented setup process

Version 10.8 | 25 August 2026

- NEW: The dashboard and admin panel are now available in 5 languages — English, Spanish, French, German and Arabic — covering every screen, button, form and notification your team sees
- NEW: Language switcher in the dashboard header, next to the light/dark toggle — the interface changes instantly, and anything half-typed into a form is kept
- NEW: Full right-to-left support for Arabic — the sidebar moves to the right and the whole layout, tables, menus and drawers mirror, with an Arabic font applied automatically
- NEW: Every user picks their own dashboard language and it is saved to their account, so it follows them to any device they sign in on — also available under Settings → Preferences
- NEW: Add your own language without touching code — copy the English translation folder, translate it, and add one line of config; anything left untranslated falls back to English, so a partly finished language still works
- IMPROVEMENT: Phone numbers, email addresses and API keys stay left-to-right inside Arabic text so they remain readable
- IMPROVEMENT: Around 2,000 lines of unused code removed from the frontend, and only the language in use is sent to the browser, so extra languages don't slow the dashboard down

Version 10.7 | 24 August 2026

- FIX: Sarvam AI agents could take up to 15 seconds to answer a short reply — saying "hello" or "yes" while the agent was still speaking is now answered in about a quarter of a second
- IMPROVEMENT: Sarvam agents now use Sarvam's realtime speech recognition, which returns words while the caller is still speaking — agents fall back automatically if your Sarvam account isn't enabled for it
- IMPROVEMENT: Interruptions now work from what the caller actually said rather than raw sound level — coughs, line noise and background TV no longer cut the agent off, while brief replies like "okay" or "haan" let it finish and are still answered afterwards
- FIX: On SIP calls, Sarvam replies longer than about 10 seconds could have a chunk silently cut out of the middle — buffered speech is no longer discarded, and playback no longer drifts out of sync on long calls
- IMPROVEMENT: Faster first word from Sarvam agents, and the voice connection is restored in the background after an interruption
- NEW: Optional voice-tuning settings for noise rejection and interruption sensitivity, so agents can be tuned to your own phone lines
- NEW: Sarvam preflight check to confirm your Sarvam account is set up correctly for voice calling before you place a call
- FIX: The backend environment template (.env.example) was missing from the v10.5 and v10.6 downloads — restored, and now documents every setting the backend reads

Version 10.6 | 24 August 2026

- FIX: HubSpot connect could fail with "code challenge parameter is missing" — HubSpot's OAuth now includes PKCE (RFC 7636) on the authorization flow
- IMPROVEMENT: Google Sheets and Google Calendar connect buttons now show a clear message instead of erroring when those integrations aren't configured on the server
- FIX: Setup script could write the wrong value into GOOGLE_CLIENT_SECRET during interactive reconfiguration

Version 10.5 | 17 August 2026

- NEW: Sarvam AI models updated — speech-to-text to Saaras v3, text-to-speech to Bulbul v3, chat to Sarvam's current 105B model; agents on deprecated models migrate automatically
- FIX: Sarvam AI voice could cut off mid-word on SIP calls — fixed, along with related causes of the agent going silent mid-reply
- IMPROVEMENT: Unified turn-taking across the Classic and Sarvam engines — brief responses like "yes" or "hmm" no longer interrupt the agent
- FIX: Deepgram Voice Agent calls could disconnect and fail to reconnect mid-call
- FIX: SIP calls could stay stuck as "In Conversation" with inflated duration after the caller hung up — hangup detection is now reliable, and duration is measured from answer time
- NEW: AI agents can now end the call themselves once the conversation is naturally over, on all three voice engines
- NEW: Idle-call detection — a connected call with no speech for 5 minutes now ends automatically

Version 10.0 | 5 July 2026

- NEW: Per-agent Voice Engine selector — pick Classic pipeline, Deepgram Voice Agent, or Sarvam AI for each agent (Agent → Voice)
- NEW: Sarvam AI voice engine for native Indian languages (Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, Indian English) — fully self-contained (Saaras STT + Sarvam-M LLM + Bulbul TTS), needs only a single Sarvam API key, per-agent language + voice with in-editor voice preview, works on both Twilio and SIP/Asterisk and supports appointment booking
- NEW: Deepgram Voice Agent engine (low latency) — one connection handles speech recognition, the LLM, and voice synthesis with server-side turn-taking for roughly 0.5–0.8s voice-to-voice responses; keeps your selected ElevenLabs voice; needs no OpenRouter key (Deepgram-managed LLM); native function-calling appointment booking
- NEW: Custom contact fields (personalized calls) — add any lead fields (e.g. {{appointment_date}}, {{service_type}}) from the Add/Edit Lead dialog or by CSV import, then reference them as merge tags in the Opening Message and System Prompt; filled per call so the AI knows each lead's details the whole conversation (works on classic and Deepgram Voice Agent engines, Twilio and SIP)
- NEW: Completely redesigned marketing landing page — bold full-screen sections that alternate your brand primary color with a clean light/dark base, an animated background grid, geometric parallax shapes, watermark typography, smooth scroll-reveal motion, feature spotlights with live UI mockups, full dark-mode support, and an admin-toggleable self-hosting section
- NEW: Latency instrumentation — every conversation turn logs queue wait, LLM first-token, and first-audio timings so you can measure real response times
- IMPROVEMENT: Classic pipeline latency tune-up — smarter end-of-speech detection and turn-taking, ElevenLabs Flash v2.5 for non-English voices, and per-call knowledge-base caching; typical responses 300–900ms faster (about 2× faster overall)
- FIX: Registration-based SIP inbound calls addressed by the trunk username are now routed to the correct agent
- FIX: Speech transcripts were processed twice per result due to a duplicate Deepgram event handler — fixed on both the Twilio and SIP call paths
- NEW: SIP Debug panel on the SIP Trunks page — a Debug button opens Health Checks and Live Logs tabs so you can diagnose SIP/Asterisk problems from the dashboard without SSH
- NEW: Health Checks — one-click diagnostics for Asterisk ARI connection, PJSIP config written and linked, Asterisk CLI access, and per-trunk endpoint / registration / provider reachability, each failed check with a plain-English fix hint
- NEW: Live Logs — real-time stream of SIP events (outbound/inbound calls, trunk rejections with the SIP cause, originate failures, bridge/recording issues, config reloads), auto-refreshing while open
- NEW: SIP scanner protection — auto-detects and temporarily blocks source IPs that repeatedly dial unknown numbers (silent hangup), tunable via SIP_SCAN_BAN_THRESHOLD / SIP_SCAN_WINDOW_MINUTES / SIP_SCAN_BAN_MINUTES
- PRIVACY: SIP debug logs are per-user — each user sees only their own trunk and call events; superadmins see everything
- FIX: Brand primary color now applies site-wide — the Primary Color set in Admin → Branding drives the public landing page, buttons, accents, and highlights (previously saved but the landing page stayed purple)
- FIX: No stock-logo flash — the default logo and favicon are blank so the original IntelliCall branding never appears for a moment before your uploaded assets load
- FIX: Landing page header no longer changes height while the logo loads (logo renders in a fixed-size box)
- IMPROVEMENT: Reset Branding restores the original IntelliCall logos and favicon (kept under /images/originals/) in one click

Version 9.7 | 23 June 2026

- FIX: Email sending issue fixed.

Version 9.6 | 14 June 2026

- FIX: Twilio inbound webhook URL on the Phone Numbers page now displays correctly.

Version 9.5 | 11 June 2026

- NEW: n8n integration — stream all platform events (leads, calls, campaigns, appointments) into n8n workflows with per-event toggles and a send-test button
- NEW: 3 ready-made n8n workflow templates included — instant call on form submit, call log to Google Sheets, qualified lead to CRM + notify
- NEW: Personal API keys — generate, regenerate, and revoke a long-lived key (stored hashed, shown once) and call the REST API via the X-API-Key header from n8n, Zapier, or scripts
- NEW: Instant AI call endpoint (POST /api/calls/dial) — dial one number with a chosen agent right now, no campaign needed; auto-creates the lead and respects plan call limits
- NEW: Built-in API reference in the dashboard — 12 documented endpoints with parameters and copy-ready curl examples
- IMPROVEMENT: Integrations tab redesigned with sidebar navigation, per-connector logos, and live connection-status indicators
- IMPROVEMENT: OAuth redirects now auto-open the matching integration card

Version 9.0 | 11 June 2026

- NEW: WhatsApp integration powered by the Meta WhatsApp Cloud API — connect with your own Meta credentials, no platform-level app setup required
- NEW: Real-time WhatsApp notifications for lead created, lead qualified, call completed, appointment booked / canceled, campaign completed, and inbound / outbound calls — with per-event toggles
- NEW: Automatic WhatsApp appointment reminders sent to the client's number before every appointment (Meta-approved message templates, configurable lead time from 15 minutes to 24 hours)
- NEW: Reschedule-safe reminders — moving an appointment re-arms the reminder for the new time; cancellations stop it automatically; duplicate sends are impossible by design
- NEW: Timezone-aware reminder formatting — dates and times follow each account's timezone and 12/24-hour preference
- NEW: Built-in test buttons — send a sample alert or your reminder template to your own WhatsApp to verify setup before clients ever receive a message
- IMPROVEMENT: WhatsApp added to the Settings → Integrations tab and the landing page integrations showcase

Version 8.0 | 24 May 2026

- NEW: Dedicated Integrations system with unified OAuth-based connector management
- NEW: HubSpot CRM integration with automatic contact sync and phone-based upsert support
- NEW: Google Calendar integration with real-time appointment sync, updates, and cancellations
- NEW: Slack notifications for lead, call, appointment, and campaign events
- NEW: Admin → Integrations panel to manage OAuth Client IDs, Secrets, and Callback URLs directly from dashboard
- NEW: Edit appointments from the Appointments page (date, time, duration, client info, notes)
- NEW: Integrations section added to the landing page showcasing available connectors
- IMPROVEMENT: Google Sheets integration moved to unified Settings → Integrations section
- FIX: Removed Twilio configuration requirement for SIP Trunk users

Version 7.4, 7.5 | 12 April 2026

- IMPROVEMENT: Updated all packages to the latest stable versions
- NEW: Configure contact email, social media links and headquarters address from Admin Settings. (shown in landing page, contact, privacy, terms pages)

Version 7.2 | 6 April 2026

- NEW: Import Knowledge Base directly from website URL
- NEW: Add and manage testimonials from Admin Settings
- IMPROVEMENT: Revamped landing page UI/UX

Version 7.1 | 29 March 2026

- NEW: Added support for CAD (Canadian Dollar) currency
- NEW: Call direction filters (Inbound & Outbound) in Agent Templates
- FIX: Resolved issue where appointment booking language defaulted to English

Version 7.0 | 21 March 2026

- NEW: Ready-to-use AI Agent Templates (sales, support, booking, outreach & more)
- NEW: Pre-configured system prompts, opening messages, and settings for faster agent creation
- NEW: Template categories for easier discovery (Sales, Support, Scheduling, Healthcare, etc.)
- NEW: Multi-language templates to support global use cases
- IMPROVEMENT: Create AI agents in seconds without writing prompts
- IMPROVEMENT: Reduced onboarding friction for new users

Version 6.5 | 18 March 2026

- FIX: DodoPayments redirect to localhost issue fixed

Version 6.4 | 17 March 2026

- NEW: Added 10 new voice languages for AI agents: Dutch, Hebrew, Italian, Japanese, Korean, Portuguese, Portuguese (Brazil), Russian, Tamil, Urdu

Version 6.3 | 17 March 2026

- IMPROVEMENT: Enhanced plan call limits to support both Unlimited Calls/month and Limited Calls/month
- IMPROVEMENT: Improved landing page UI/UX for mobile devices
- IMPROVEMENT: Branding app name included in test call message
- IMPROVEMENT: Improved setup and deployment scripts
- FIX: Fixed Google OAuth login issue
- FIX: Fixed favicon not updating issue for custom branding

Version 6.0 | 11 March 2026

- IMPROVEMENT: Enhanced user interruption handling during calls. AI agents now listen and respond more naturally when users interrupt while speaking
- NEW: Branding settings added in Admin Panel (customize app name, primary color, logos, and favicon)
- IMPROVEMENT: Agents page UI/UX fully revamped with a cleaner interface and new create/edit agent drawer
- NEW: "Check for Updates" feature in Admin Settings to easily check for the latest IntelliCall AI version
- GENERAL: Performance optimizations and stability improvements

Version 5.1 | 9 March 2026

- FIX: Resolved known issues related to SIP Trunk integrations and improved stability across SIP providers
- IMPROVEMENT: Enhanced UI/UX across multiple sections for a smoother user experience
- NEW: Customization and branding options added in Admin Panel
- NEW: Web Installer introduced for easy product installation and one-click updates

Version 5.0 | 5 March 2026

- AI Appointment Booking
- Knowledge-Powered AI Agents (RAG-based knowledge system)
- Multilingual Voice Agents with auto language detection (English, Arabic, French, Hindi, German, Spanish)
- Email notifications for important call events (leads, calls, appointments)
- Smart Auto Hang-Up protection for inbound and outbound call duration limits
- Admin-configurable trial plan limits
- Landing page dark mode support
- Updated dependencies and packages
- Backend refactoring and stability improvements

Version 4.1 | 1 March 2026

- BUG FIX: Setup scripts and documentation in v4 incorrectly referenced .env instead of .env.local, causing API requests to default to localhost