AI photo review for Safesite safety actions

Overview

A photo in Safesite is evidence for one reported thing, and it is usually looked at exactly twice. Someone photographs the hazard when they raise it, someone photographs the fix when they close it, and after that the image sits in a resolved record that nobody reopens. But the frame holds far more than the hazard it was taken for: the ladder behind the guardrail, the extension cord across the walkway, the housekeeping in the corner that no one filed an action against. RapidEye reads the photos already sitting on your Safesite projects and returns a finding for each of those, so the evidence you collected does more work than the one line it was filed under.

How it works with Safesite

Safesite calls the things people record safety actions, and each kind carries photos for its own reason. RapidEye reads them where they already are.

• Hazard photos. A recorded hazard carries pictures taken in real time or picked from the phone's gallery, plus a location and a responsible team. RapidEye returns what else is in those frames and, where a location has been photographed before, whether the same condition keeps coming back.
• Resolution photos. Closing a hazard needs a comment or a photo, and supporting photos of the fix can be added. RapidEye reads the resolution photo against the original one and reports when the frame does not show the condition actually cleared.
• Observation photos. Positive and negative observations both carry images, a location and a description. Negative observations are the closest thing on site to a running condition log, and they read well as a series.
• Inspection photos. Templates can require a photo on a negative answer, or on any answer. Those images are tied to a specific question, which makes them the cleanest input there is: RapidEye can say whether the photo supports the answer that was given.
• Incident report photos. The scene as it was found. A finding here is about what the photo shows around the incident, which is the part a later investigation always wants and rarely has.

Each finding names the project, the location, what the issue is, how serious it looks, the photo with the region marked, an earlier photo of the same place where one exists, and a suggested next step. It lands in the RapidEye dashboard and, if you want, a Slack channel or an inbox for the safety team. Nothing is recorded, resolved or archived in Safesite by RapidEye. If the standard you want checked is not a safety one, a client's site condition spec, a scaffold handover, an equipment condition report, or a video walkthrough of a floor instead of stills, that is a setup conversation rather than a limit: RapidEye reads whatever documentation the site already produces and reports against the standard you hold it to.

How to connect Safesite to RapidEye

1. Pick the projects. Projects are the unit in Safesite. Personnel, locations, campaigns and every safety action hang off one, so the projects you pick are the scope of the integration.
2. Keep photographing the same way. Hazards, observations, inspections and incident reports already carry images, in real time or from the gallery. Nobody installs a second app or changes how they report. If a project needs a fixed walk of set angles that no safety action covers today, RapidEye can send a capture link that opens on any phone.
3. Add RapidEye in Organization Personnel. Organization, then Personnel, then Add Personnel, with the projects selected on the invite. The modal has the details.

What a Safesite record actually holds

Checked against Safesite's help centre, September 2026.

Safesite is a safety management platform built around the phone in a worker's pocket, and it is unusually complete about what a person on site can file: a hazard, a safety observation, an inspection or audit, an incident report, a safety meeting. According to Safesite's help centre documentation (safesitehq.com), all of these start the same way, from an action card on the home screen or the plus icon that opens a New Safety Action screen, and every one of them is attached to a project and, in most cases, a location. For photos, the documentation is specific: standard image files such as JPEG, GIF and PNG are supported on a hazard or an observation, while PDFs and other document formats are not, and images can be captured in real time or picked from the gallery.

The hazard, and the assessment attached to it

A recorded hazard is more structured than most tools of this kind manage. Alongside the description, photos, location and responsible team, the person raising it fills in a hazard and risk assessment, and Safesite computes from it.

Closing it has its own rules. Unresolved hazards show as action cards under the Hazards tab, only personnel on the responsible team can resolve one, resolution requires a comment or a photo, and additional supporting photos can be added from the camera icon at the top of the hazard. Who is allowed to resolve and archive hazards is itself an organization setting. All of which means a closed hazard in Safesite is a genuinely well-evidenced record, which is exactly why it is worth reading again.

What an inspection answer can be made to require

Inspections run from templates, either Safesite's own library or ones your organization builds, and a location has to be chosen before the questions start. The questions are criteria checks, and the answer set is where the photo requirement lives.

That last rule is the reason this integration is useful rather than redundant. A completed inspection is a fixed record of one person's judgment, question by question, and it cannot be revised when someone later notices the photo shows something the answer did not. RapidEye reads the attached photo and reports what is in it, which gives the safety lead the option of raising a new hazard against the same location instead of arguing with a locked record.

Does the reading still work when the site has no signal?

Yes, once the phone syncs. Safesite documents that the app works offline in full after the initial login sync, with the exception of a few online-only functions such as exporting PDF reports, and that anything recorded offline does not reach the admin dashboard until the device is back online; its own guidance is to stay in the app for a few minutes until the sync banner reports complete. RapidEye reads photos once they have arrived in the account, so a hazard raised in a basement, a tunnel or a rural site is read exactly like one raised with full bars. Safesite's product pages also list the app in English, Spanish, French and Portuguese, which is worth knowing here for a reason people miss: a photo needs no translation, so a finding read from a Spanish-language observation reads the same to the office as any other.

Safesite integration FAQ