One widget. Live chat, voice, video, and an AI agent that can actually do things.

What Laleo Actually Is

Laleo.chat isn't a chat widget with a "call us" button bolted on. It's a unified live-engagement platform where a single embedded surface handles four conversation modes — typed chat, voice call, video call, and AI-agent autopilot — over the same session, with the same visitor identity, and the same backend. A visitor lands on a customer's site, the launcher renders in the corner, and within one click they're connected: to a real agent if one is online, to a Laleo AI agent that can take real actions on the site owner's behalf if not, and to voicemail with audio recording if nobody is around. No app install. No app store. No phone dialing.

The reason that matters is that "chat widget" as a category has been static for a decade. Intercom, Drift, LiveChat, Tidio — they all ship a textbox with maybe a Calendly link and a Zendesk handoff. Laleo collapses that whole stack: the visitor monitor, the helpdesk inbox, the click-to-call dialer, the video meeting room, the AI agent, and the analytics dashboard are one product, one widget, one account.

The "Voice + Video + Text" Single-Widget Pitch

The visible heart of the product is a small launcher that lives in the bottom-right of every page. When the visitor opens it, they see a chat thread. When they request a call — or when a human agent inside the customer's dashboard initiates one — the widget upgrades the same session to a WebRTC voice call. If either side toggles a camera on, it upgrades again to video, full-screen, picture-in-picture, with camera-flip and zoom controls on mobile. All of it runs inside the visitor's browser via RTCPeerConnection. No plugin. No "click here to download Zoom." No phone number dialed. No code shared.

The engineering ramifications are larger than they sound. Because the call is bound to an active chat session, the agent already has the visitor's page history, scroll depth, current section, the path they came from, UTM source, and the last thing they clicked. Picking up a call doesn't start from "How can I help you?" — it starts from "I see you're on the pricing page and you've been reading the enterprise tier for three minutes." That context bridge is what makes voice and video usable as a first-touch channel rather than a last-resort escalation.

A Telephony Backbone Behind a Modern Browser Front-End

Laleo is built by ABSG Tech, which is a VoIP shop first and a SaaS company second. That genetics shows up where it matters. Laleo doesn't pretend WebRTC is a complete answer to "talk to customers." It treats the browser widget as one termination point on a real PBX. Two things follow:

SIP peer trunking into an existing PBX

Customers who already own a phone system — FreePBX, FusionPBX, 3CX, a hosted Asterisk, a carrier SBC — can SIP-trunk Laleo as a peer. Calls from the website widget ring desk phones and softphones already on their PBX. Outbound calls to the visitor's widget originate from extensions they already own. There's no rip-and-replace, no porting numbers, and no asking agents to learn a new dialer. Laleo sits adjacent to the system that already works.

PSTN forwarding as a fallback

For SMBs without a PBX, every Laleo plan — including the free tier — includes PSTN forwarding so a website call can ring an ordinary cellphone or office line. The starter tier includes 60 minutes of PSTN-forwarded time per month, which covers the case where the customer wants the convenience of website voice without standing up phone infrastructure. Forwarding is metered honestly: minutes are visible in the account-usage view, and overages bill against the plan.

That dual posture — peer-trunk for the businesses that have a phone system, PSTN-forward for the ones that don't — is the telecom-shop signature on a product that otherwise looks like pure SaaS.

The LLM Agent — and Why MCP Changes the Equation

Laleo's AI agent isn't a scripted FAQ bot. It's a large-language-model agent with tool access through MCP — the Model Context Protocol — and the tools are the same set of operations the dashboard exposes: list widgets, fetch live visitors, read conversation history, send a message, pull analytics, generate an install snippet, set up a webhook, set campaign aliases. Because MCP is the integration surface, the agent isn't bolted on top of the platform; it operates the platform.

The practical consequence is that "AI chat" stops being a deflection layer. When a visitor asks "Can you book me a call with someone in sales for Thursday at 2?", the agent doesn't say "I'll let them know." It creates a conversation link, addresses the right campaign, sends the link to the right human, and writes back "I've put you through to Maria — she'll have your context loaded." When an agent or admin asks Laleo "Show me everyone who's been on the pricing page for more than thirty seconds in the last hour," the AI calls the live-visitors tool, filters, and answers. The MCP surface is the agent's hands.

CRM and Workflow Integration

Every visitor interaction in Laleo emits structured events: page views, scroll-depth, section dwell, clicks with selector tags, UTM tracking on first-landing, message-sent, call-requested, call-accepted, call-completed, voicemail-left, video-on, video-off. Those events are available through the platform's webhook system — fire to any HTTPS endpoint with a signed payload. That's enough to feed any CRM: HubSpot contact updates, Salesforce lead creation, Zoho or Pipedrive deal stages, or a custom Postgres ingest.

For accounts that don't want to write integration code, the same data is exposed through the visitor activity, visitor messages, and call-history MCP tools so an AI assistant or no-code platform (n8n, Zapier, Make) can read it on demand. Conversation links — short-lived URLs that drop a CRM user directly into a visitor's session view — close the loop: sales rep clicks a link in their CRM, lands inside the Laleo conversation context, takes the call.

Open API + MCP — Built for Developers

Underneath the MCP surface is a documented OpenAPI specification. Every operation an agent can perform through MCP corresponds to a versioned REST endpoint with stable schemas. That matters for shops that want to skip the conversational layer and integrate directly: ingest events into a data warehouse, drive widgets from a CMS, programmatically provision new widgets for a multi-tenant agency, pull analytics into Looker or Metabase. The pairing — OpenAPI for traditional code, MCP for AI agents and assistants — means the same primitives are reachable from both directions.

WordPress Plugin — The One-Click Path

The fastest path to live on a website is the WordPress plugin. A site owner signs up at laleo.chat with an email, installs the plugin, pastes a widget ID, and the widget renders on every page. No theme edits, no header injection, no learning JavaScript. The plugin handles the asset version pinning so widgets self-update in the field — a stale plugin still gets fresh widget code without a WordPress plugin update.

For sites not on WordPress, the install-snippet endpoint returns the exact <script> tag to paste into any HTML template, headless CMS, or framework partial.

Pricing — Free Until You Add Humans

This is the line that separates Laleo from the rest of the category. All product features are free. Voice, video, the AI agent, the analytics dashboard, webhooks, MCP, the OpenAPI, the WordPress plugin, conversation links, campaign reporting, visitor presence, voicemail with audio recording — none of it has a paywall.

You pay for seat licenses, and only for seat licenses. A "seat" is a logged-in human agent who picks up calls and chats inside the operator dashboard. One-person shops who let the AI handle everything pay zero. Two-person shops where one human is online pay for one seat. Agencies running fifteen client widgets with a five-person inbox pay for five seats. Calls handled by the AI don't count against seats. Visitors don't count. Widgets don't count. Conversations don't count. Messages don't count.

PSTN forwarding minutes are the only metered overage, billed transparently against the included 60-minute starter allowance, with higher allowances on paid-seat plans.

The Tech Stack

The widget is vanilla modern JavaScript — no React runtime shipped to the visitor, deliberately, so the launcher loads small and adds no measurable page weight. Real-time signaling is a single WebSocket per session, plus HTTP fallback for environments where WebSockets are blocked. Media is RTCPeerConnection with adaptive ICE configuration fetched per-session — that's how the engine handles symmetric NAT and corporate firewalls without forcing a TURN server config on the customer. Voicemail recording uses the browser MediaRecorder API and posts back as a blob. Asset versioning is pinned and auto-refreshing.

On the backend, it's a Node/TypeScript service for the signaling layer, an LLM gateway for the AI agent (model-agnostic — same widget runs against Claude, GPT-4 class, or self-hosted), a Postgres + Redis store for sessions and presence, an MCP server, an OpenAPI gateway, and a SIP/Kamailio layer for telephony bridging. Hosting is multi-region for low call setup latency.

Who It's For

Laleo is built for: small and mid-sized businesses that want a single answer to "how do we talk to people on our website"; WordPress site owners who want voice and video without building a phone system; agencies running multiple client sites who want one dashboard and one bill; VoIP shops who want to extend their PBX onto their customer's website; and developers who want a programmable, MCP-native customer engagement layer they don't have to host.

It is not built for billion-dollar enterprise call-center replacement — Genesys and NICE still rule that floor — and it doesn't pretend to be. It's built for the long tail of websites that today have a "Contact" page and nothing more.