AI concierge
The Copiara concierge
One page, kept current, describing exactly what the AI in Copiara does, what it is allowed to touch, and what it will not do. Every AI claim we make anywhere else links back here.
Status: in build. Copiara is pre-launch, so everything on this page is being built for the first pilots rather than running in production today. When something ships, we date it here.
What it is
A chat assistant docked beside the buyer portal, built on Persona, an open-source embeddable chat widget. It is not a separate product, and it does not keep its own copy of your data.
How it gets its answers
The concierge calls the same authenticated endpoints the screens call. Tenancy and permissions are not re-implemented for the AI. If a buyer is not allowed to see a price, the concierge cannot see it either. Every tool call runs as that signed-in buyer.
What a buyer can ask it to do
Search the catalog
In buildFind parts by description, specification, or SKU.
Cross-reference a part number
In buildResolve a competitor or OEM number to your SKU, with the confidence it has.
Check contract price and stock
In buildThat buyer's negotiated price, their volume breaks, and current availability.
Add to an order or start a quote
In buildStage lines on a draft order, or open a quote for the rep desk.
Place an order
In buildOff by default. When a distributor turns it on, the same approval threshold applies as in the portal, enforced by the server and not by the prompt.
What a rep can ask it to do
Summarize an account
In buildRecent orders, open quotes, credit position, and what is waiting on someone.
Draft a counter reply
In buildThe server computes a counter that clears the margin floor. The model writes the sentence around it. It does not invent prices.
Guardrails
- The approval threshold is enforced server-side at order submit. There is no separate path for the concierge.
- Tool calls are scoped to contract parts for that buyer.
- Model output is sanitized before it is rendered.
- Every tool call writes an audit row naming the user it ran as.
- Per-tenant spend budgets are checked before the model is called, so an assistant cannot run up a bill unattended.
The model
Gemini, Flash tier, through Vertex AI is the default. The provider layer is model-agnostic, so a tenant can be moved to a different model without changing the tools. Usage is metered per tenant, so a distributor can see what the assistant costs rather than finding out at the end of the month.
Your data
Your catalog, pricing, and order data are used to answer your buyers' questions. They are not used to train a model.
What it will not do
- It does not negotiate for you. A counter comes from your rep or from your rules, not from the model.
- It cannot see another company's orders, prices, or quotes.
- It cannot approve its own order.
- It does not replace your rep. It handles the lookup so the rep can handle the relationship.