The problem
Manuals and training videos miss what experts actually do: the small adjustments, the order of steps, the judgement calls made in real conditions.

Capture real world skills. Build the data foundation for physical AI.
Why Vigil
Manuals and training videos miss what experts actually do: the small adjustments, the order of steps, the judgement calls made in real conditions.
Real demonstrations of real work, captured where it happens: on busy sites, in changing conditions, from the person doing the job.
A physical AI data platform that captures skilled work from the worker’s perspective and is being built to prepare demonstrations for robot learning and searchable team knowledge.
Language models learned from libraries of written text. The skill in a pair of hands was never written down.
Vigil captures what manuals miss.
How it works
Vigil clips to the brim of a hard hat or cap. No screens, no menus, nothing to hold.
Start and stop with your voice or the single button. The camera, depth and motion sensors record the task from your point of view.
Recordings wait safely on the device and upload over wireless internet or LTE as soon as a connection is available.
Each demonstration becomes searchable task episodes with timestamps, ready to review, reuse and export.
Vigil clips to your cap, leaving both hands free for the task.


Two ways to take part
Capture it for your own team, or contribute your skills and get paid for them.
For organisations
For individuals

Physical AI
Robots need examples of work in the real world. Vigil captures skilled people performing tasks from their own perspective and is building a pipeline to prepare those demonstrations for robot learning.
Video from the wearer’s point of view, with depth and head motion captured alongside: how the task looks to the person doing it.
Recordings become episodes and steps with timestamps, so a robot learning team gets clean examples, not hours of raw footage.
Real sites, real tools, real variation. The messy detail that lab demonstrations miss and robots most need.
Export curated episodes as datasets for training robot policies and physical AI models, on your terms.
Private by design
Vigil can run entirely inside your environment. Recordings, transcripts, annotations and search indexes stay on infrastructure you own.
Connect the AI services your company already approves. Nothing is sent to ours.
If a model is unavailable, recordings simply wait. Nothing is routed elsewhere.
Annotations keep their evidence, versions and reviewer decisions.
Datasets leave your environment only when you export them.

Under the hood
A compact module that clips to the brim of a hard hat or cap, carrying the sensors that matter for capturing skilled work.

The film
Bring Vigil to your team and start capturing the real world expertise that can power robot learning and team knowledge.