Relay in use, from brief to result.

Recorded workflows, from the first brief to the reviewed output.

Browse all 10 films

Task examples, checked before training

Define a production-calculation task, generate examples, check the answers and select the same dataset in ModelForge.

Ready

Recorded in the Axeron / TeamAI desktop app. Relay is the planned name. Full application view. Example packets; waits are shortened.

In this film

Download options for this demo

Explore the films

10 recorded demos

The next question, with the work in view.

How does this become a repeatable process?

Save the roles, tasks and writing standards with a team. Supply a new packet for each assignment, review the plan, follow execution, and check the actual files. The creator film shows a saved writing standard used with a fresh brief. The banking film reuses a saved team.

How much time could a team save?

Use net time after review and corrections. These are planning inputs, not measured customer savings.

Hours per week = jobs × (current minutes − assisted minutes) ÷ 60. A pilot should also track errors, rework, model costs and escalation rates.

Where does human control appear?

The recordings show plan review before execution, visible agent activity, an in-app document edit with retained history, checkpoint recovery, and a paused marketplace deployment. These are specific observed controls. The enterprise permission ceiling and private company catalogue require deployment-level verification.

Can this be a marketplace for our company?

The founder’s enterprise model is a private catalogue of prepared teams for employees and portfolio companies. Administrators govern access to public imports. Teams must inherit the adopting employee’s authority, with traceable actions and intervention. The film labels this as the enterprise vision; the desktop experiments did not prove complete enforcement.

What could a creator sell?

A specialist method can support a service: a clear brief, reusable instructions, test examples, a useful deliverable, and ongoing maintenance. A team can also be prepared for marketplace review. These demos do not establish approval, customer demand, payouts or earnings.

How do Synthetic Data and ModelForge work together?

Define a task, generate examples and inspect their actual answers. The recorded workflow checks production calculations, then selects the same sixteen-row dataset in ModelForge. Model training, held-out evaluation and deployment are separate steps; a dataset handoff alone does not establish model quality.

How were the films made?

These edits use continuous OBS recordings of actual desktop interaction, with the complete application window in a fixed frame and shortened waits. Step labels and captions sit outside the application. No close-ups or camera pans are added. Access credentials are hidden where they appear in the deployment view. Narration was produced with ElevenLabs. Captioned and clean versions use the same recorded actions. No generated interface, staged success screen or synthetic pointer was substituted.

The banking film follows the successful corrected attempt after an earlier budget-related failure. The creator pack completed through checkpoint recovery. Its training example is a separate earlier evaluation. Final presentation edits removed broken logo references from the written outputs. The marketplace clone was paused and tailored; the three industry outputs came from separate teams.

Website analytics

We use aggregate website and video statistics to improve these demos. This page does not ask for your name or email address. How Vercel handles analytics.

Share this demo

Copy this link into your message or social post.