What is CRSTA?
The short definition
CRSTA is the AI conversation agent at the core of the CRSTBL platform. When a visitor talks to CRSTBL on your website, business page, kiosk, or trade show booth, they’re talking to CRSTA.
CRSTA does three things in every conversation:
- Reads the question in the visitor’s own language.
- Answers from the business information you’ve given it — menu, catalog, service list, policies, FAQs, documents.
- Labels the exchange as a stored context: what was asked, how it was answered, what it revealed about customer intent.
Those three things repeat across every visit, and each one adds to what CRSTA knows about your business and what your customers are looking for.
Why “CRSTA”? The story behind the name
CRSTBL is pronounced Crystal Ball. The name was chosen for a specific reason.
Every business owner wants a crystal ball — a way to see what customers actually want before they leave the site, close the app, or walk out the door. Traditional website analytics can tell you what customers did: which pages they visited, how long they stayed, where they bounced. It can’t tell you what they were trying to do, or what they’d have bought if you’d met them where they were.
Conversations reveal that. Every question a visitor asks is a signal. Read them well and you’re close to a crystal ball.
CRSTA is the piece of the platform that reads those signals. The agent in the crystal ball.
How CRSTA reads a question
When a visitor types (or speaks) a question, CRSTA doesn’t match it against a script or a fixed keyword list. It interprets the sentence — the way a knowledgeable employee would — and looks up the answer against the business information you’ve given it.
The mechanic has four steps:
Parse the question in natural language. Handle typos, colloquialisms, industry slang, multi-part sentences, and follow-ups that reference earlier turns (“and what about the price?”).
Retrieve the relevant portion of your business information. If the question is about menu items, look at the menu. If it’s about hours, look at hours. If it’s about a product spec, look at the catalog.
Reply in the visitor’s own language, in the voice defined by the business’s Character Identity. If the answer is unknown, say so clearly rather than guessing.
Save the exchange as a labeled context: which topic, which intent type, which product or service, whether it led to a next step. This is the raw material for conversation analytics.
CRSTA answering real customer questions.
Watch CRSTA deployed as a restaurant menu concierge — the same architecture powers every CRSTBL agent.
CRSTA, Conversation Agents, and Character Identity
Three terms come up around CRSTA. Here’s how they fit together:
- CRSTA is the underlying AI conversation agent — the software.
- Conversation Agent is the deployed unit. Your restaurant’s website concierge is one Conversation Agent. Your kiosk in the lobby might be another. Each Conversation Agent runs on CRSTA.
- Character Identity is the reusable brand persona that lives on top of CRSTA — the voice, tone, and style the agent uses. One Character Identity can drive multiple Conversation Agents, so your website concierge and your kiosk agent sound like the same brand.
The analogy: CRSTA is the actor. Character Identity is the costume and script direction. A Conversation Agent is a specific performance in a specific venue.
What makes CRSTA different
Compared to a general-purpose LLM (like the one behind ChatGPT), CRSTA is grounded in a specific business’s information. It won’t make up a menu item you don’t have or invent a return policy you didn’t write.
Compared to a traditional chatbot, CRSTA handles unscripted questions in real language and remembers context across turns of a conversation. (For the full comparison, see CRSTBL vs Traditional Chatbots.)
Compared to a website search bar, CRSTA answers the question rather than returning a list of links that might contain the answer.
What CRSTA doesn’t do
A few honest limits worth naming:
- CRSTA doesn’t replace human staff. It handles routine, repetitive, and discovery-style conversations so staff can spend time on the ones that need judgment, negotiation, or relationship-building.
- CRSTA doesn’t invent information. If a customer asks a question the business hasn’t provided the answer to, CRSTA logs the gap rather than guessing.
- CRSTA doesn’t work well without organized data. The quality of CRSTA’s answers depends on the quality of the business information provided. Onboarding is largely about getting that data right.
Frequently asked questions
How do you pronounce CRSTA?
CRSTA is pronounced “Crysta” — the same way you’d say the first two syllables of CRSTBL (“Crystal Ball”). It’s a compressed spelling, not an acronym you spell out letter by letter.
Is CRSTA a large language model?
CRSTA uses large language models under the hood, but it’s not a raw LLM. It’s a conversation agent built on top of language-model technology, grounded in your business’s specific information and shaped by a Character Identity you control.
Can I have more than one CRSTA?
Each Conversation Agent runs on CRSTA. A single business can deploy multiple Conversation Agents — for example, one on the website and one on a lobby kiosk — and share a Character Identity across them so they sound like the same brand.
Does CRSTA work in languages other than English?
Yes. CRSTA supports multilingual conversations. This matters for restaurants and retailers serving diverse customer bases, and for enterprise deployments in multi-region businesses.
See conversational discovery in action.
CRSTBL turns your website, menu, or catalog into a conversation that guides visitors to what they came for — and shows you what they were looking for.