Design Labels with AI — Talk to LabelInn in Plain English

Most "AI features" in software-as-a-service are an autocomplete sticker. They describe what you typed back to you in a chat panel and leave you to do the work. LabelInn's AI assistant is built differently: it actually executes against the platform's tool surface. You ask, it acts.

If you say "create a 4×6 inch shipping label template with the order ID and a QR code linking to the tracking URL," it creates the template. If you say "print 20 copies on the warehouse-A Zebra," it queues the print. The output of the request is not text describing what it would do. The output is the action.

Built on Gemini, gated by a tenant credit pool

The assistant runs on Google's Gemini 2.5 Flash. Every customer's account has a monthly AI credit pool — 50 credits on the free plan, 150 on Starter, 300 on Pro, 1,000 on Enterprise. A turn is a credit; long, multi-tool requests are still a credit. Credits reset on the billing cycle.

Power users who want their own provider rate-limit and bill — or who want to use a higher-tier Gemini model — can paste their own Google API key into Settings → API Keys → Google AI. From that point, the assistant uses the customer's key, charges the customer's quota, and stops counting against the LabelInn credit pool.

This matters because AI cost is unpredictable at scale and we don't want a hot operator's prompt to surprise a customer's CFO. The pool gives a hard cap. The BYO key gives uncapped power-user freedom. Customers pick.

What the assistant can actually do today

The assistant has direct access to a large, structured surface of LabelInn capabilities. Each capability is a typed tool the model can call, with a clear schema for inputs and outputs. The model decides which tools to call to satisfy a request; the tools execute against the live system, not against a sandbox.

Tools exist for:

• Template authoring — create, clone, modify, delete templates; add elements (text, barcode, image, QR, rectangle, line); set fonts, sizes, positions; publish snapshots.
• Print operations — submit print jobs; query status; cancel queued jobs; reprint failed jobs.
• Fleet management — enumerate printers; check status and supply levels; route jobs by printer role.
• Data binding — import a CSV; bind elements to columns; preview the rendered output.
• Integration queries — fetch open orders from connected marketplaces; query cargo carrier rates.
• Audit + provenance — surface the audit log entries for a date range or a specific print event.

The total surface is large — over a hundred tools across a dozen modules. The model sees them as a typed API and picks the right combination for each request.

Prompts that work today

A few examples drawn from real customer usage:

• "Create a new shipping label, 4×6 inches, with the order ID at the top, the customer name and address centered, and a QR code in the bottom right pointing to the tracking URL."
• "What printers are online right now? Show their queue depth and supply levels."
• "Reprint the last three failed jobs on whichever ColorWorks printer is online."
• "Show me every label printed on Line 3 in March, with the template snapshot that was active at the time."
• "Clone the shipping-pro template, change the title to '{order_id} — {customer_name}', and render a preview."

Each of these produces an action, not a paragraph of text. The model returns a brief confirmation of what it did, with the relevant identifiers (template ID, job ID, printer name) so the user can follow up.

Privacy posture

Prompts and label-design context are sent to Gemini only for the duration of the individual request. The platform does not pool prompts for training, does not retain them beyond the request lifecycle, and does not share them with any third party other than the LLM provider for that request. The same is true if you bring your own Google API key — your prompts go to Google under your own Google account's terms, not under LabelInn's.

For regulated customers (pharma, medical device) on FDA Part 11 tenants: the AI assistant is opt-in per tenant. By default, AI calls do not execute write actions against templates linked to controlled SKUs; the assistant can read but a human signs off on the write. The opt-in policy is set by the tenant owner and audited like every other compliance setting.

What's on the roadmap

Two related capabilities are in development for Q3:

• A second LLM provider option (Claude). Designed, infrastructure in place, currently in internal validation. Customers preferring Anthropic for cost, safety, or vendor-diversification reasons will be able to switch with a setting flip.
• AI-authored automation — beyond template editing, the assistant will be able to propose rule changes and data transforms. Proposed automations go into a pending-signatures queue so a human approves before they ship into production.

Neither is generally available today. The current Gemini-based assistant is.