Train an AI bot on your business

A chatbot is only as useful as what it knows. A bot that hasn’t been taught your refund policy will guess, hedge, or invent one — and the customer ends up arguing with the human you wanted to protect. A bot that has never seen your product catalog will recommend products you don’t sell. A bot trained on the open internet will sound confident about your business while saying things that are not true.

The fix is to ground the bot in your own content. That step — getting your content into a form the bot can search through and quote from — is where most chatbot projects either stall for months or end up depending on engineering time you don’t have. NebulaHex collapses the whole pipeline into a single screen with an Add Source button. You upload, paste, or point. We do the rest.

A real knowledge base is messy. Some content is in Google Docs that someone exported to PDF. Some lives on your help center. Some is in a deck the team uses for onboarding. Some is in a video your founder recorded last year. NebulaHex handles all of those shapes without asking you to flatten everything into one format first.

You can mix as many sources as you like inside one bot. Most growing businesses end up with a help center crawl, two or three policy PDFs, a product CSV, and a YouTube playlist of explainers — five to ten sources covering tens or hundreds of pages of content, all answerable through a single chat surface.

You don’t need to understand the mechanics to use the product, but a sentence of plain-language detail helps. When you add a source, NebulaHex splits it into smaller passages — chunks — each sized to be a self-contained idea. The split is automatic and tuned so that an answer to a typical customer question lands inside one or two chunks rather than being scattered across a whole document.

Each chunk is then indexed in a way that lets the bot search the entire knowledge base by meaning, not just by exact keyword match. When a customer asks a question, the bot finds the chunks most relevant to what they’re actually asking — not what they happened to type — and uses those chunks to write its reply.

A static knowledge base goes stale fast. You change your refund window. You launch a new product line. You update the shipping policy because a courier rate changed. The bot needs to know.

Every URL source has an Auto-sync control with four settings: Off, Daily, Weekly, and Monthly. Pick the cadence that matches how often the source actually changes. The crawler re-runs on schedule, picks up the new content, and updates the bot’s knowledge automatically. If you can’t wait — you just updated a page and want the change live now — the Sync Now button runs the crawl immediately.

For files and pasted text, freshness works differently: replace the source. Delete the old one, upload the new one, the bot is updated. Most teams build a habit around this — the document that the bot reads is the same document the team edits, so updating the bot is one extra step on top of an edit you were going to make anyway.

The knowledge dashboard is built so a non-technical team member can run it confidently. Each source has a status pill — Pending, Processing, Ready, or Error — so you always know what state it’s in. A Processing source shows a spinner. A source that failed for any reason shows the error message inline so you can fix it without digging into logs.

Good chatbots are auditable. If a customer ever gets an answer that surprises you, scan the chunks for that source and you can see exactly what content produced it. The dashboard’s reader view shows the extracted content the way the bot sees it — the same passages, the same structure.

The knowledge base is one half of how the bot works. The other half is live data — the ability to look up things that change in real time, mid-conversation.

Live data flows through the integrations catalog: look up real orders in Shopify, real tickets in your helpdesk, real records in your CRM. The Live Data picker in the dashboard also handles direct connections — Google Sheets, product catalogs, custom API endpoints — for the cases where you don’t need a full integration but you do need the bot to read something that updates regularly.

Static knowledge tells the bot how your business works. Live data tells the bot what’s happening in your business right now. Together, they cover the substantial majority of questions a customer-facing chatbot ever needs to answer.

You can build this in-house. Many teams have. We’ve talked to enough of them to know how it usually goes: it takes a quarter, it costs more than the build estimate suggested, the maintenance burden eats into the next quarter, and it ends up looking suspiciously like what we ship — except yours has more bugs and your engineers are tired.

NebulaHex sits at the intersection: zero engineering, knowledge-grounded answers, and source management any team member can run. You feed it anything. It learns everything. Auto-sync keeps the bot up to date as your content changes, the way it was meant to be from the start.

Your knowledge content is encrypted at rest and in transit. Access inside your team is role-based: Owner, Admin, Editor, and Viewer roles let you decide who in the workspace can edit knowledge versus who can only see it. Sensitive actions require confirmation.

We do not use your content to train AI models, and we do not share it across customers. When you delete a knowledge source, the chunks the bot was drawing on are removed and the bot stops referencing that content in subsequent conversations. The same applies when you delete a bot or close your account. See our privacy policy and terms.

The whole point of this product is that the gap between “we have content” and “we have a bot answering customers from that content” should be measured in minutes, not months. You don’t need to hire an engineer. You don’t need to learn what chunking is.