Each request goes to the right place, decided before anything runs.
Step 1 of 4: Known work
Every request: answer, hand over or decline.
Known work is answered at once, new work goes to your AI, and work that breaks a rule stops.
Confidence and autonomy. It runs On every request.
The requests Duplex knows stop costing a model call.
On a busy day the same requests come back again and again. Duplex answers the ones it has learned, so your AI is called for what is new.
Step 1 of 4: A repeat
The fortieth address change of the day arrives.
Your rules are checked before anything answers.
Every request meets your rules first. Only then can Duplex answer from what it has learned or brief your AI, so a request that breaks a rule reaches neither.
Step 1 of 4: Rules first
A request arrives and meets your rules before anything else.
Fewer model calls, and no attack let through.
On 3,000 customer-operations decisions, Duplex cut model calls by 90.3% and had the highest accuracy in the test. Against 300 paraphrased attacks, none got through, with the fewest false refusals of any approach that blocked them all.
Step 1 of 5: Model calls
On 3,000 decisions, Duplex answered the requests it knew, and model calls fell by 90.3%.
Decides whether to answer, hand over or decline.
For every request, Duplex answers from what it learned, hands the work to your model or a tool, or declines, and can show why.
- Every request: answer, hand over or declineKnown work is answered at once, new work goes to your AI, and work that breaks a rule stops.
- The examples behind a decisionAsk why, and Duplex shows the confirmed examples behind a decision.
- Every handed-over step, closed onceEach step runs exactly once and none is lost, so a payment or a deployment never happens twice.
See it on your own work.
From your own assistant to every agent your company runs, a pilot puts Duplex on your own work, with the measures agreed before it begins.