Support, catalog and search on models sized for the work.

Most calls in a commerce pipeline are narrow. Konic runs them on right-sized models, so the AI bill stops growing faster than revenue.

The challenge for online retailers

AI can carry support, catalog and search. The bill is what holds it back.

Online retailers run into the same walls:

AI costs grow faster than revenue.
A single support conversation can trigger a long chain of model calls, each one billed at frontier rates.
Most of those calls are narrow.
Routing a ticket or reading a product attribute does not need the largest model available.
Agents that cannot act just add a step.
A support agent is only useful if it can reach orders, refunds and shipping.
Switching models feels like a gamble.
Quality has to hold on real traffic before anyone will move a production agent.

Konic for this industry

Support agents that act.
Duo-1 runs the tool calls behind a support agent: orders, refunds and shipping.
Catalogs that fill themselves.
Cero-1 extracts product attributes from seller images and text, so listings arrive complete.
A switch you can measure.
Run in shadow mode on real traffic, compare quality side by side, then move traffic over in stages.

Illustrative rollout

An example path, not a customer story. Your scope and timeline are agreed with you.

E-commerce

Routing narrow decisions to small models removes most of the expensive calls.

Keep reading