New skills, learned on its own.

Routine work nobody scripted becomes a skill Duplex runs itself, where it is safe to try.

Reusable skills. It runs On request, once your team switches it on.

100%Of tasks completed with the model switched off, and every learned skill removable. See the result →

Set a goal and the limits. Duplex works out the skill, and proves it.

Your team gives Duplex a goal, such as setting up a new client’s workspace, with the actions it may take and a way to check success.

A task Duplex has learned stops costing a model call.

Once the skill is proven, every new client after the first gets a workspace set up by Duplex itself, and your AI is free for work that is new.

The first new client’s workspace is worked out step by step, and the result is checked.

Every task completed with the model switched off.

Duplex was tested on four suites over ten runs, with no examples, labels or help. It completed 100% of tasks, refused every unsafe shortcut and could remove every skill it learned.

The model was switched off for the whole test: a single model call would have failed the run.

It only tries what you allow.

Duplex learns new skills where it is safe to try and where success can be checked. An action outside the limits you set is refused before it runs, and the skill is still learned.

Creating the client’s project board is within its limits, so Duplex may try it.

Learns new skills on its own.

Where it is safe to try and success can be checked, Duplex learns a new skill, tests it before use and can retire it.

Next: A wrong skill withdrawn, with what it madeSee it →

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.

Start a pilot →All 94 capabilities

New skills, learned on its own.

Set a goal and the limits. Duplex works out the skill, and proves it.

Your team gives Duplex a goal, such as setting up a new client’s workspace, with the actions it may take and a way to check success.

Duplex tries the task safely, using only the actions it was given, and sees what each one does.

A try that fails the check shows it what does not work, so the next try gets closer.

Once a try passes the check, Duplex has a proven skill, ready for the next new client.

The next new client’s workspace is set up by the skill itself, with no model call.

A task Duplex has learned stops costing a model call.

Every task completed with the model switched off.

It only tries what you allow.

Learns new skills on its own.

See it on your own work.