Familiar work, answered. New work, briefed.

Duplex answers what it has learned in milliseconds, briefs your AI for what is new, and learns from every confirmed result.

One request, every way it can go.

Familiar work skips the model call, so it comes back faster and costs less.

A decision Duplex has learned is answered in ≈97 ms, where a model call takes ≈2,231 ms.

Each work area sets its own standard, and Duplex learns only from results that meet it.

When a result meets the test your team chose, it is confirmed, and Duplex learns from it.

What Duplex learns from, what it leaves out, and how you change it.

Duplex learns from pretraining material, confirmed corrections, its decisions and their outcomes.

Duplex knows when it knows, and hands over the rest.

A repeat quote is one Duplex has learned, so it answers at once, with no model call.