AI agent that handles customer questions and brings in your team when needed

The idea

Relayd isn't an AI chatbot. It's an AI agent: a system that doesn't just answer customer questions, but also performs actions, keeps track of conversations across multiple channels and knows when it's better to bring in a human.

The agent talks to your customers via chat, WhatsApp, email, Telegram, Instagram, Messenger and, recently, by phone as well. The team sees everything in one inbox and can take over any conversation with a single click. The AI handles the routine work, people deal with what deserves attention.

Not a chatbot, but an agent

The difference is in the scope. A chatbot picks an answer from a list. An agent can investigate, decide and act:

• Answer product questions based on your own catalog and knowledge base
• Look up orders, start returns, check refund status
• Schedule meetings via Cal.com or Google Calendar
• Create tickets in Linear when something can't be resolved within the channel
• Warmly hand over a conversation, including full context, so the customer doesn't have to repeat anything

For every answer, the agent calculates a confidence score. Below a set threshold, or for sensitive questions, it automatically brings in a team member. That team member receives the complete conversation and steps in smoothly.

Human takeover

This is the most important part of the platform. The AI isn't a replacement for your team, but a filter. Any team member can disable the agent for a specific conversation at any time, have it generate suggested replies, and hand it back to the AI once things become routine.

Saved replies, mentions, internal notes and conversation analysis turn the inbox into a real workspace, not a passive log.

Voice channel

Recently, the agent can make and receive calls too. Through an integration with ElevenLabs for voice and Twilio for telephony, the agent answers incoming calls, holds conversations in a natural Dutch voice and can warmly transfer to a team member or external number.

Voice conversations land in the same inbox as the text channels, with transcription and recording playback. Outbound calling is possible too: start a conversation from a contact card, or have the agent call back automatically when that's part of a flow.

Flows: build your own workflows

Alongside free conversation, you can define flows: visual workflows with conditions, actions and handoffs. The FlowMatcher automatically recognizes when an incoming message fits a flow and starts it. Think of quote requests, return handling or new-lead qualification.

The FlowSuggester analyzes real conversations and proposes flows based on the patterns it sees. That way you keep building out the platform based on what your customers actually ask.

E-commerce integrations

The agent connects directly to your webshop. Drivers for Shopify, WooCommerce, Lightspeed eCom and Medusa are built in. Each driver supports whatever the platform can handle: orders, catalog, customer history, gift cards, returns, subscriptions, restock notifications, reviews and promotions.

The customer asks: "Where is my package?" The agent looks it up in Shopify, checks the track-and-trace and answers right away. No extra step, no ticket.

White-label portals

For agencies and partners there's a white-label layer. Each partner runs its own portal with its own domain, its own brand colors, its own email domain via Resend, its own Mollie key and its own pricing. That partner's customers never see Relayd, they see the agency's brand.

Including partner administration, audit log, login-as impersonation for support, per-portal credit caps and automatic monthly portal invoices.

What else it can do

• Multi-channel inbox - All channels on one screen, with media exchange: photos, documents, voice notes
• Trained on your own content - The agent learns from your website, FAQ, product information and manuals. Answers are in line with what you would say
• Proactive activation - The widget starts a conversation based on visitor behavior: lingering on pricing, scroll drop-off, returning visits
• Knowledge gaps - The platform flags questions the agent didn't have a good answer for, so you can fill your knowledge base where it matters
• 90+ languages - One agent, every language. Answers in the customer's language
• Two-pool credits - Separate budgets for text and voice, so voice usage doesn't quietly eat up your text budget
• Privacy by design - EU hosting, GDPR-proof

Architecture

Backend on Laravel 12 with PostgreSQL and Redis. Frontend in React 19 via Inertia.js. Realtime through Laravel Reverb for the inbox, live chat and voice events. Background tasks such as AI analyses, media processing and website scraping run via queues with Horizon monitoring.

Multi-tenant from day one. Each tenant has its own knowledge base, its own widget configuration, its own integrations and its own team. The messaging layer is uniform: a single ChatMessage model with attachments, kept separate from the channel. Specific clients (Meta GraphClient, TelegramClient, WhatsApp Cloud API, ElevenLabs/Twilio for voice) translate incoming messages into that model. Adding a new channel means writing a client, not rebuilding the whole inbox.

Embeddable widget

The website widget is a standalone JavaScript bundle that works on any site with a single script line. Loaded async, non-blocking, custom themeable per tenant. The widget on this portfolio is a live example.

The bundle is deliberately kept small and only loads after first paint, so it has no impact on Core Web Vitals.

Result

The platform is live at relayd.chat . Companies use it to scale their customer contact without extra headcount: the AI absorbs the volume, the team stays in control. The combination of agent behavior, human takeover and native e-commerce integrations makes it more than a chat widget and less than a full call-center platform: exactly where most companies need it.