Assistants
An assistant that acts, not one that chats.
Nexora builds AI assistants, chatbots and voice agents that carry out work rather than describing it. The assistant inside Extrack, a product of ours, can take 29 different actions on its owner's behalf — live at myextrack.com. The difference between that and a chat box is engineering, not prompting.
- AI chatbots and lead qualification
- Voice agents
- AI workflows and automation
- MCP servers and tool-calling pipelines
- WhatsApp and enquiry integrations
What separates an assistant from a chatbot?
Whether it can do anything. A chatbot answers; an assistant books, files, updates, refunds, chases. Extrack's assistant has 29 actions available to it, each one wired into the product properly, and that wiring is the entire job. The conversation part is the easy half.
What is an MCP server, and do I need one?
MCP is the protocol that lets an assistant call your systems as tools rather than guessing at them. You need one when you want an assistant to work against your real data instead of a copy of it. Nexora has shipped MCP servers and tool-calling pipelines in production, including Extrack's.
Can an assistant qualify leads before a person sees them?
That is one of the most common builds. It asks the questions a salesperson would ask first, writes the answers where the team already looks, and hands over a conversation that has already started. WhatsApp is usually where it happens.
How long does an assistant take to build?
Three to four weeks. That band comes from having done it: Extrack went from idea to first paying customers in 15 days, assistant included.
How do you keep an assistant from inventing things?
By giving it tools instead of latitude. An assistant with 29 defined actions can do 29 things and report honestly when the answer is none of them; one with a prompt and a hope will fill the gap. The engineering is in the tool definitions and what happens when a call fails.
What it is built with
TypeScript, React and Next.js, on AWS — Lambda, SES and the rest. MCP servers for tool-calling, the Claude API where a model is doing the reasoning, and Playwright end-to-end tests over the paths that matter. WhyCosmos runs on that shape and carries more than 2,400 tests.