Minaya.ai
An AI website assistant that answers from your own content, captures leads on intent, and acts inside your systems through MCP
Visit Live SiteAbout this project
Minaya.ai is TieTech's own AI product: an assistant that sits on a company's website and handles the conversations that would otherwise go unanswered after hours or get lost in a contact form. It answers visitor questions from the business's own knowledge base site content, PDFs, FAQs in the brand's voice and in whatever language the visitor writes in. It watches for buying intent and, when it appears, captures name, email, and phone and emails the team immediately. And through the Model Context Protocol it reaches into real business systems to book appointments, check orders, and update inventory inside the conversation, instead of handing the visitor off to another page. The engineering problem behind a product like this isn't the chat UI it's grounding answers in a customer's specific content so the assistant doesn't invent things, keeping per-tenant knowledge and transcripts isolated, encrypting customer-supplied model keys at rest, and running an embeddable widget that loads fast on someone else's site without fighting their CSS. It ships as a zero-code install: configure the widget, add knowledge sources, drop in a script tag. This site runs it.
Three Jobs, One Widget
Minaya covers the work that usually falls between a static FAQ page and a support inbox: answering questions around the clock, spotting a lead before it leaves, and actually completing the task the visitor came to do.
- Answers 24/7 from the customer's own site content, PDFs, and FAQs
- Detects buying intent mid-conversation and emails captured leads to the team
- Books, checks, and updates inside connected systems via MCP
- Replies in whatever language the visitor writes in
It Doesn't Just Answer. It Acts.
Most website assistants can only talk. Through the Model Context Protocol, Minaya reaches into the systems that hold the real answers and performs the work in the conversation. A visitor asks where order #4471 is; Minaya queries Shopify live, reports that it shipped yesterday and is out for delivery in Denver, and when the visitor asks to cancel, it recognises the order is close to delivery, sends a one-time code to verify identity, and completes the cancellation and refund without a human touching it.
- Live order lookup against Shopify, not a cached or scraped status
- Cancellations, refunds, and updates executed in the source system
- Step-up verification with a one-time code before irreversible actions
- The whole exchange stays in the chat, no handoff to another page
Order Status, Stock Levels, Payment History, Live From The Source
Connecting a tool is a toggle, not an integration project. Shopify, Stripe, Zapier, GitHub, Linear and Jira attach through MCP servers, and from the moment one is switched on the assistant can read and act against it. Because the data is fetched live at conversation time, an answer about stock or payment status reflects the system right now rather than whenever a sync last ran.
- Shopify, Stripe, Zapier, GitHub, Linear and Jira connect as MCP servers
- Enabled per tool with a toggle, scoped to what that server exposes
- Queried live during the conversation, so nothing goes stale between syncs
- Available on Pro plans, where real system access is warranted
It Only Answers From What You Give It
The failure mode for a website assistant is confident invention. Minaya answers from an explicit knowledge base the customer controls paste text, crawl a URL, upload a PDF so responses stay anchored to the business's real policies, catalogue, and documentation rather than whatever the model happens to believe.
- Knowledge added by pasted text, crawled URL, or uploaded PDF
- Answers grounded in that source material, not general model knowledge
- Per-tenant isolation keeps one customer's content out of another's replies
- Sources editable after launch as policies and catalogues change
This project may have been built for a third-party client who outsourced development to TieTech Innovations, or for a direct client. It is shown to demonstrate our engineering work only see the disclaimer in the footer for details.