Skip to content

Try Sight in a minute

See a photo check work, from your own code, before you train anything. You make a test key on your dashboard and send one of our sample photos to Sight's sample model, which finds hard hats. The answer comes back in a few seconds: a verdict, and the evidence for it.

1. Make a test key

  1. Sign in, or create a free account, at sight.raku.so.
  2. On your dashboard, under Try Sight in a minute, choose Make my test key.
  3. Copy the key straight away. It is shown once: Sight keeps only a fingerprint of it, so nobody, Raku included, can show it to you again. If you lose it, make a new one.

2. Check a sample photo

The quickest way is the command line: npx @raku-technologies/sight login, paste the key, then npx @raku-technologies/sight try. Or call the API yourself:

Run this, with your test key in place of YOUR_TEST_KEY (the dashboard shows it with your key already in it, while the key is on screen):

curl https://sightapi.raku.so/checks/photos \
  -H 'X-Api-Key: YOUR_TEST_KEY' \
  -H 'Content-Type: application/json' \
  -d '{"template":"sample-hard-hats","external_job_id":"my-first-check","slot":"site_photo","image_url":"https://sight.raku.so/samples/crew-in-hard-hats.jpg","mode":"sync"}'

The photo shows a crew in hard hats, so the answer's verdict is pass. The answer also lists the check that decided it ("Hard hats seen on site") and the hard hats the model found, with how sure it was of each.

Now send the other sample, workers without hard hats, as a new job:

curl https://sightapi.raku.so/checks/photos \
  -H 'X-Api-Key: YOUR_TEST_KEY' \
  -H 'Content-Type: application/json' \
  -d '{"template":"sample-hard-hats","external_job_id":"my-second-check","slot":"site_photo","image_url":"https://sight.raku.so/samples/no-hard-hats.png","mode":"sync"}'

This time the verdict is fail, and the check says what it needed: a photo of the crew on site with hard hats in it.

What the sample checks

The sample model finds hard hats. The sample check passes when it sees at least one hard hat in the site photo; it does not count people or check each head. Your own checks can ask for much more: see Writing templates.

3. Try your own photo

Any photo Sight can download over https works: change image_url to a link to your photo (a presigned link from your own storage is fine), and give each job its own external_job_id. To send the photo's bytes instead of a link, see Sending photos.

From TypeScript, the SDK does the same with template: "sample-hard-hats" and slot: "site_photo".

What a test key can do
May Send photos for checks, and read their results. Nothing else: no predictions, no templates, no results webhook.
Model Sight's sample model, which finds hard hats.
Photos Up to 50 a day (they start again at midnight UTC), counted in your free account's 1,000 photos a month.
Lasts 30 days. Make a new one on your dashboard at any time; making one stops the one you had at once.
Whose results Yours only. Your checks are kept in a project of your own called Try Sight, which only you can see.

A key that has stopped working says why, and the dashboard shows how many of today's photos it has used.

Next

  • Tutorial: train your own model in the console, on photos of your own kind of work.
  • Photo checks: how verdicts, evidence and "needs a person" work.
  • For AI agents: how an AI agent should call Sight.

Sample photos: "FEMA Public Assistance Officers with County Official at Dam" by George Armstrong, FEMA, public domain; workers without hard hats from Voxel51's hard-hat-detection set, CC0.