GUARDHELM PRIVACY NOTICE
Version 2026-11-22a
Effective 22 November 2026
This notice explains what GuardHelm collects about the people who use it, who can see it, and how long it is kept. It is written for guards. Office staff who use the GuardHelm console are covered too, in their own section near the end.
GuardHelm is run by WASAW LLC, doing business as GuardHelm. In this notice "we" means WASAW LLC. "Your employer" means the security company you work for, which chose to use GuardHelm and decides how parts of it are set up.
Questions, or a request about your data: hello@guardhelm.com
THE SHORT VERSION
- New firms start with On duty only. That limits the background points GuardHelm keeps, not when your phone can take a reading. A phone with background location allowed can still take, queue and send points off duty. Ask your employer which setting applies to you.
- Clock-ins, patrol visits, signed reports and receipts are work records. A correction sits beside the original, not in its place. There is no routine expiry for signed work records. Other records, such as messages, have different periods below.
- Your employer can choose to record your calls in the app, with the office and with the GuardHelm assistant. Recording is off unless it does, and a recorded call starts with a voice saying so. You can talk with the assistant only after you say yes. Your voice goes live to OpenAI, Google or Cloudflare to be answered. Transcription is automatic, but the saved text can be incomplete or missing.
- If your employer turns on shift checks, GuardHelm can contact you about your shifts, for example when you may be late, by a notification or a call from the assistant. See LOCATION for when a shift check may use your position.
- Office staff can see work records within their roles and sites. Your employer's clients can see reports and receipts shared with them. Your employer can choose to share your name or on-duty path with a client. Neither is on by default. A report can still identify you by what it says or shows. Your employer can also give a client a file it downloaded.
- If your employer connects an outside program, the program can receive information about your work even if you never use it. A schedule connection, a token for the office and an assistant connected by one person give different access. An assistant connected by one person can read message and report text, documents and pay, and what it reads goes to the company that runs it. See WHO CAN SEE YOUR DATA.
- We do not sell your data. There is no advertising or analytics code in GuardHelm.
LOCATION
When it is recorded
The guard app has a background tracker. Your employer has this on unless it turns it off. While you are signed in and your phone allows background location, the tracker can take readings even when the app is closed. Clocking out changes how often it asks, not necessarily whether it runs. It can keep readings on the phone and send them when it has a connection, even if GuardHelm does not keep them in your location history.
While the tracker runs, it asks for a point after about 25 metres of movement on shift and 100 metres off shift. It also asks for a check about every 5 minutes on shift or 15 minutes off shift. Your phone may take fewer readings. A still iPhone off shift may send nothing.
Your employer chooses what GuardHelm may keep and show:
- On duty only: new firms start here. When a background point arrives, GuardHelm checks the time it was taken against the signed duty records it has then. It does not add a point outside that duty time to the tracking history.
- On request: the same limit on background points. In addition, office staff who cover the whole firm can ask for one position or ask who is near a site. The request does not ring the phone or send a notification. The current phone apps do not answer these requests. After two minutes the requester sees no answer, not a position or proof you were away.
- Always: GuardHelm can keep points taken off duty as well as on duty. Firms using GuardHelm before these choices were introduced kept Always unless they changed it.
If you clock out offline, GuardHelm can still treat you as on duty and accept later points until it receives your clock-out. Under On duty only or On request, the map uses the corrected duty times and can hide points it had shown. Correcting the map does not delete points already accepted. They can remain in storage until the usual cleanup, and it cannot undo what someone already saw.
Separately from background tracking, the app asks for one reading when you clock in, clock out, start or end a break, and at a patrol stop that asks for a photo. Turning off the background tracker does not turn off those checks.
What a point contains
Latitude and longitude, how accurate your phone thinks the reading is, the time, your battery level, whether your phone flagged the location as faked, and what caused the reading: moving, a timed check, or a change to your location settings. Speed, heading, altitude and what you were doing are not sent. We do not turn your position into an address.
How you can tell
The app shows the tracker status. When it is running, Android can show a persistent location notification and iPhone can show its location indicator. Those signs tell you about the phone, not whether GuardHelm kept a reading or the office saw it.
Who controls it
- Your employer can turn tracking off for the whole company, or for you alone. Each change needs a reason and is recorded with the name of the person who made it.
- We can pause tracking for a whole company, and lift that pause. We have no switch for a single guard.
- The app has no on/off switch. You can turn location off in your phone's Settings, or allow it only while the app is open. Background recording then stops. Clocking in and out still works, and your pay does not depend on your location.
- To select On request or Always, your employer must confirm that guards agreed in writing. GuardHelm records who made that choice and when. The employer keeps its paperwork. That entry is not your signature, and allowing location on your phone is not a written agreement.
- When tracking is paused, GuardHelm refuses new tracking points from the pause. An offline phone may keep sampling until it hears the change. Signing out asks the phone to stop. It can finish once GuardHelm disables the tracking credential even if the phone has not confirmed that it stopped sampling. Office accounts do not use this background tracker. See OFFICE STAFF USING THE CONSOLE for watch positions.
Who can see your location
- Supervisors, dispatchers, admins and owners who cover every site can see an available on-duty position. With Always they can also see the latest available position off duty and the trail of off-duty points kept while Always was on. GuardHelm logs who viewed off-duty positions. On duty only and On request do not give the office a continuous off-duty map or trail.
- An On request search is logged with the requester and guard. Its result is for the requester, not the whole office or your location history. The result window closes ten minutes after the request expires. The next cleanup removes the request row, not the access log.
- Staff limited to certain sites see no off-duty guard on the live map. They can see an on-duty position at a site they cover and earlier evidence from a shift there, including positions recorded with a clock-in, clock-out or break.
- If the phone is silent during a shift or location is turned off or approximate, the office can get an alert. A missing point is not proof that you were absent.
- An office incident packet can show the phone's available positions around the incident during the shift. The time of a reading and the time you reported an event are separate facts.
- For a call you start, if GuardHelm's records show you clocked in, the AI service may get a position check: whether you are on post, the distance and direction to it, the reading's age and accuracy, whether it was marked as simulated and whether tracking is on. The check does not send your phone's coordinates. A site address is separate. If your clock-out is still on your phone, the server may still treat you as on duty for this check. If it sees you off duty, the service may get a tracking setting instead. Calls the assistant places do not get the position check.
- A client can learn where you worked from a report that names a site or shows checkpoint locations and scan times. Your employer can also share points from your on-duty path. This is not access to the office's live map. See WHO CAN SEE YOUR DATA for what may be shared.
Before a shift
If your employer turns on shift checks and their use of location, a check can use an available position from an hour before a shift until you clock in, or until the shift ends if you do not. Only Always allows this automatic use of off-duty location. It is used to decide whether to contact you, for example to ask whether you are on your way. Without a usable position the system can still ask you.
The shift check screen shows your office what you answered. It does not show the position the check used or its travel estimate. Office staff can still see your location as described above. An owner or admin covering every site can get the check reference, distance and any estimated travel time in the firm's records download.
If a travel estimate is made, the check keeps the distance, any estimated travel time, the age and accuracy of the reading, the travel mode and the decision. The check does not store the coordinates in that estimate. A background location record is separate.
No map or traffic service is used for this today. If one is added, this notice will change to name it first.
How long it is kept
- Background location points are due for deletion 90 days after they were recorded. A nightly job works on whole days, so deletion is not at the exact 90-day mark.
- The live map normally draws on about five days of recent points. Older points can still be in the live database if archiving is delayed. The office can also see when the phone was last heard from without seeing a position.
- Points your phone could not send are dropped from the phone after 3 days, or oldest first once more than 5,000 are waiting.
- Some points are kept longer. A point that set off an alert your employer has to review, such as leaving your site during a shift, is kept with that alert. Points under a legal hold are kept until the hold ends. A position GuardHelm accepts with a punch or a photo stop is kept with that record. A punch can be accepted even when its position is not stored.
WORK RECORDS
Your phone signs records such as punches, patrol visits, incident reports, follow-up notes, notes for the next guard at your post and entries in site logs. A photo or voice note filed with a report belongs to that report. A receipt records the server's answer. It does not mean a person has read the report.
Once receipt timestamping is deployed, GuardHelm automatically asks FreeTSA and Sigstore for timestamps on signed receipts, including older ones. There is no setting for your firm to turn this on or off. We send these independent services a digital fingerprint of a batch, not the receipts, their contents or direct identifiers for people, firms or work records. The request comes from GuardHelm, not your phone. FreeTSA says it logs the sending network address to check for abuse and does not keep the fingerprints.
Someone who has a signed receipt can match it to the fingerprint for a batch of one. For a larger batch they can match it using the receipt and its proof. A timestamp can show that the signed bytes existed by that time. It does not prove when or where you worked.
GuardHelm keeps the fingerprints, batch details and replies with no set end date, even after a firm's records are erased. A fingerprint already waiting can still be sent after erasure.
In the guard app and the office console, a signed record stays as it was filed. Nobody can edit or delete a work record. A mistake is fixed by adding a correction beside the original with the name of whoever made it. Your office can add a corrected version of a report without changing your words. This is the rule for work in the apps. The separate procedure for closing and erasing a whole firm is described below.
A guard leaving the company does not delete the firm's copy. The rules for ordinary work records do not mean that the service keeps every type of data forever.
Notes you leave at a post can be read by guards assigned there and by office staff with access. A patrol visit can record how you checked in, such as a scan, a photo or a manual entry. A tag scan can include the tag identifier. It does not identify your phone by its hardware serial number.
Your employer may ask you to keep a site log, for example to record a visitor's name, company, badge or vehicle plate, who they came to see, or a photo. Log answers are stored encrypted. Office staff covering the site can read or export entries. Your employer can allow entries from a log in a client report. Ask the firm keeping the log about an entry that concerns you, even if you are not a GuardHelm user.
No end date has been set for keeping work records. There is no separate routine expiry for site log entries. If your employer closes its whole GuardHelm account, that is a separate process, not an individual guard deleting a shift. Before you leave, ask your employer for any copy you need.
Report text, follow-up notes, saved voice transcripts and punch positions are encrypted on our servers. Opening them needs a key held outside the database. The assistant's separate write-up of a filed report does not have that extra encryption. Incident photos and voice recordings are stored as uploaded in private Cloudflare storage, without that extra encryption. The signed record contains a fingerprint for each file. A receipt for the record is not itself proof that the file finished uploading.
Your employer can set a pay rate for a shift. GuardHelm works out hours from your punches and shows a pay estimate when it has the information it needs. It does not pay your wages.
THE CAMERA AND PHOTOS
The camera is used to:
- scan the setup code your office gives you
- scan patrol checkpoint tags
- take a photo at a patrol stop that asks for one
- add a photo to an incident report or a site log that asks for one
The app takes a photo when you press the shutter. A draft photo can stay on the phone. When you file a photo, the app prepares a copy without camera location metadata and signs a record that identifies the file by its fingerprint. The separate upload may not have reached the office when it accepts the record.
Photos stay with the report. A client may see incident and patrol photos if the office has enabled photos for that client. Otherwise the report can show counts. A site log shared with a client shows photo counts, not the photos themselves.
THE MICROPHONE, VOICE NOTES AND TRANSCRIPTS
The microphone is used during a call in the app, and when you tap Record to make a voice note in an incident report or a message thread. A voice note stops when you stop it, or after two minutes. If your employer has call recording on, most calls in the app are recorded too, as described under CALLS.
Cloudflare runs the AI model that turns voice notes into text for GuardHelm. Making a note can send it for transcription without a separate question about AI. The text is marked as automatic and may be wrong. It never replaces what you write.
- Once a report and its recording reach GuardHelm, the app can request this transcript automatically. Your office can request it too. The recording and saved transcript stay with the report.
- In a message thread, the text is made after you send the note. Both follow the rules for messages below.
- If you use the transcript preview before sending a report, the recording goes to GuardHelm and to Cloudflare to make the text. GuardHelm does not file the preview as a transcript of a submitted report.
- You can ask for an AI draft from the transcript. An AI model that Cloudflare runs for us writes it. Generating a draft does not file it. If you choose Use this draft, its words are saved in your unfinished report. You can edit them before signing and sending it, or send the words you chose.
This editable preview is not a second version of a filed report.
After you file an incident report or add to it, GuardHelm can ask the same AI model at Cloudflare to write an English account for your office. It gets your report text, later additions, available voice note transcripts and the kind, severity and time of the incident. Names in your words can go with them. No photo or audio is sent for this write-up. This is on by default. Your firm's owner can turn it off. Your choice about assistant calls does not control it.
The write-up can be wrong. It is not signed and never replaces your signed words. GuardHelm stores it separately as readable text, with no set expiry. Turning it off does not erase earlier write-ups. Office staff who can open the report can read the write-up beside your words. It can be in the firm's PDF and export, not in GuardHelm's client copy. A file someone has downloaded is outside GuardHelm's control.
Record your own account. Do not record other people without telling them, because some places require everyone's consent.
AUTOMATIC CHECKS
GuardHelm can ask an AI model that Cloudflare runs for us a set question about something you record. The model only answers. It never changes what you signed, never lowers a report, never decides about you and is never used to score you. A person or one of your office's existing rules acts on the answer. GuardHelm shows an answer only once answers to that question have been measured to be reliable; until then it is kept but not shown. An answer that is shown is marked as automatic and may be wrong.
- Incident reports. After you file a report, its words go to the model to ask whether it describes harm or danger and how serious it is. If it may describe harm, or be more serious than filed, the report moves up your office's list, marked as an automatic reading. Email addresses, phone and card numbers, other long numbers and map coordinates are replaced first. Names in your words can go with it.
- Assistant calls. What you say on a call with the assistant can go to the model to ask whether someone is in danger now, so your office can be alerted sooner. The same numbers are replaced first. Only an answer that someone may be in danger is kept.
- Photo stops. A photo you take at a checkpoint that asks for one can go to the model beside your office's photos of that spot, to ask whether it shows the same place. You and your office see the answer, except on a random sample your office reviews without it. Your visit counts from your signed photo either way.
- Voice notes to your office. The words of a voice note you send your office in the app can go to the model, with your next few shifts, to ask whether you mean to call off one of them, so the app can offer you a card to confirm it. Only you see that card. Nothing changes unless you tap it. The same numbers are replaced first. Neither your words nor the answer are kept.
- Report drafts. While you write an incident report, its words can go to the model to ask whether it says where, what happened and who was involved, so the app can suggest what to add before you send it. It never stops you sending, and nothing about it is kept.
- Post orders. The post orders your office writes can go to the model to suggest which licences a post needs. An office person decides whether to add each one.
Your firm's owner, or an admin who covers every site, can turn each check off for the whole firm. An owner or admin can turn photo checks off for one site. GuardHelm keeps the question, the answer and the model's version with the record it is about, not another copy of your words or photos. While a check is being tested, its answers are deleted after 90 days. Once a check is on, every answer about a record, shown or not, is kept as long as that record.
CALLS
Calls with your office
You can call your office from the guard app and the office can call you, including for a welfare check. A call can appear on your phone's call screen while locked. Whether it rings depends on the phone, permissions and network. The app checks the account addressed by the call before letting you take it.
These calls run over the internet through Cloudflare, not the ordinary telephone network. GuardHelm keeps a call record: who called whom, which shift or site the call relates to, its times, the ringing or answer events reported, and whether it was recorded. Your office can see that record. A call attempt is not proof that anyone heard it. Call records are kept for a year unless covered by a legal hold.
If an in-app ring fails, an authorized office user can look up the guard's number and choose to call it. If nobody can take your office call in the app, you may be offered an office number or a voice message. There is no automatic transfer to a carrier call. A call you choose to make through your phone's own dialer is not recorded by GuardHelm.
Calls in the app are not for emergencies. In an emergency, call 911 from your phone. On Help, dialling 911 or the site's phone number uses your phone's own dialer. Call the office is a separate in-app call where available.
Calls with the GuardHelm assistant
The assistant is a computer, not a person. You can call it from the app, and if your employer turns on shift checks it can call you, as described under SHIFT CHECKS. It answers questions about your shifts, your post and its instructions. On a call you make, it can check whether you are on post if the server's records show you clocked in. It can also describe work on your shift, such as your punches, breaks, checkpoint scans and check-ins. It can get things wrong, so check anything important with your office.
If you ask it to tell the office about an incident or a problem on shift, the assistant can send a message in your name. It can also alert the office to a problem it hears without being asked. It uses what it understood, which may differ from your words. Depending on the service answering, it may ask for confirmation first. A correction is a follow-up, not a change to the first message.
An office message from a call is not a signed incident report. Signing and filing a report in the app is a separate action. The assistant can also report a delay, call off a shift or request time off when you ask. If you ask for help or it hears possible danger, it can alert your office. A message or alert is not proof that someone has read it or is on the way.
- Before your first call with the assistant, the app asks your permission. It asks again whenever the words of that question change. You can say no and still use everything else, and change your mind later on the Help screen. A ring already prepared can still reach you after you withdraw permission. GuardHelm checks permission again before connecting you to the assistant.
- Each call begins with a computer voice saying "This call is transcribed by" and your employer's name. If your employer records calls, it says "This call is recorded and transcribed by" and your employer's name instead.
- We try to save the text of what you and the assistant said, including what you said while it was answering. A call with no words to save has no transcript. Missing speech or a failed save can leave a partial transcript or none at all.
- If your employer records calls, GuardHelm also tries to save audio of your side and the assistant's, starting no earlier than the assistant's first words. A recording can have gaps.
- On a connected assistant call, your voice goes live to the AI service answering you. That can be OpenAI or Google. On a call you make, Cloudflare models can provide a fallback. The opening call notice is a separate step. If another service takes over, it can receive the conversation so far. GuardHelm records which service answered.
- When you start an assistant call or answer one it placed, the AI service can receive your first name, employer, shifts, post, site and work activity before it hears your voice, even if setup fails. It can be told that duty is unconfirmed. Depending on the service, it can also get post orders during setup. A lookup can give the service the site's address, emergency details or support contact. Only a call you start can include the position check described under LOCATION.
- If you ask, the hours lookup gives worked time, not your pay amount. On a call asking you to stay, the service may be told whether some or all of the extra time would earn a premium. Other lookups can return unread message counts rather than message text, titles of announcements you have not confirmed, open shifts you can take, time-off requests or your own note at a post. If you report a delay, call off or request time off on a call, the service hears the details you give it.
- Anyone near your phone during the call is heard by that service too, and what they say can end up in the text.
- OpenAI can keep what it receives and says on a call for up to 30 days to check for misuse, and longer if the law requires it or to prevent harm. A second OpenAI model used for lookups can store its requests and answers. GuardHelm tries to delete those saved answers when the call ends. If that fails, they can remain under OpenAI's retention rules.
- GuardHelm does not request a resumable Google call session. Google may keep some data to check for misuse. Our call settings do not establish how long Google keeps its copies.
- For the call services described here, OpenAI and Google say they do not use calls to train their models. This is not a statement about the separate Claude or ChatGPT account your employer might connect to its schedule.
- An assistant call has a time limit. The firm's remaining minutes may end a call sooner or prevent it from starting.
- Saved transcripts and recordings are encrypted and kept with the firm's call record. After one year, routine cleanup removes the call record and saved text unless a legal hold covers them. See below for recording files.
When your employer records calls
Your employer can turn call recording on for the whole company. It is off unless your employer turns it on. Only an owner can turn it on or off, and each change is stored with who made it and when. There is no switch for a single guard. The app does not tell you when recording is turned on or off.
When recording is on, GuardHelm can record these calls and tries to transcribe them:
- calls between you and your office in the app, whoever calls, on shift or off shift
- calls with the GuardHelm assistant, whether you call it or it calls you
GuardHelm does not record:
- the automatic welfare calls, which ring with "I'm OK" and "Need help" buttons
- calls you make with your phone's own dialer, including calls to 911 or your site's phone number
An assistant check-in is different from the automated welfare ring, even if both show the same buttons. It is an assistant call and follows the firm's call recording setting.
GuardHelm records who asked you to check in, when it was due, your answer and when it reached us. The office can add a note if it confirms for you. A missed check can page the office or ring your phone. A page does not mean anyone responded.
Whether a call is recorded is decided when the call is made. Changing the switch during a call does not change that call.
A recorded call starts with a computer voice saying "This call is recorded and transcribed by", then your employer's name.
On an office call set for recording, GuardHelm waits until the call is answered and it can send the spoken notice to both sides. Recording starts with that notice. If the recorder cannot prepare and send the notice to both sides in time, the call can continue unrecorded. Sending the notice is not proof each person heard it.
Listen for the spoken recording notice on a call with the office. What the phone shows can vary with the app build and with which call screen is open.
A recording has two tracks: your side of the call, and the side of the office staff member or the assistant. After a recorded office call, GuardHelm uses a speech-to-text model at Cloudflare to transcribe each side. The text can be incomplete, wrong or unavailable. The model is set for English, so words in other languages may come out garbled. Lines are labelled by the side of the call, not the speaker: someone else's words picked up by your phone can appear under your name. GuardHelm does not make voiceprints or use recordings to recognise anyone by their voice.
Anyone near your phone on a recorded call can be recorded too, and may not have heard the sentence. Tell them the call is recorded, because some places require everyone's consent.
Saved recordings and transcripts are encrypted. Opening them needs a key held outside the database. The nightly cleanup can remove call records and saved text more than 365 days after the call began, unless a legal hold covers the call. It destroys the recording key and queues the encrypted files for deletion. File removal can take longer. Turning recording off does not erase recordings already made. See WHO CAN SEE YOUR DATA for who can open them.
Your employer can use recordings and transcripts to keep a record of what was said, look into incidents, complaints and disputes, and for legal claims. GuardHelm handles these records for your employer. We do not sell them, share them for advertising or use them to train AI models.
Some states, including New York, Connecticut and Delaware, require employers to give their own written notice of monitoring. This notice does not replace it.
SHIFT CHECKS
Your employer can turn on shift checks. They are off unless it does, and only an owner can turn them on or off. GuardHelm can ask you to confirm an upcoming shift. It can also contact you when something may be wrong, such as a late start or late relief. When it needs an answer it usually sends a notification with buttons first. If you do not answer, the assistant may call you.
The assistant may call you to:
- ask whether you are on your way, when your shift is about to start and you have not clocked in
- confirm a shift you have coming up
- ask whether you can stay on when your relief is late
- check in with you during a shift, as welfare checks do
The assistant needs your permission. A call about a late start, late relief or a welfare check can come during quiet hours, a sleep window or before the time you chose for calls. For other calls, allowing off-duty calls can lift the quiet hours and sleep window, but not your earliest call time. Repeated calls are limited. You can change call preferences in the app.
An assistant call rings like an office call. It starts with the spoken notice described under CALLS. GuardHelm tries to transcribe it and, if your employer has recording on, to record it.
Where an updated shift check accepts a staying answer on a published shift while duty is unconfirmed, the office sees that fact beside the answer. If GuardHelm cannot save your answer, contact your office directly.
Unless your post requires you to stay until relieved, you can say no to staying. An unanswered notification can be followed by a call. No or Too tired can end the request and alert the office. If duty was unconfirmed when you answered, we record the answer without closing the request or sending a new notice because of it. A staying answer does not change your punches, scheduled hours or pay. Can't make it can cancel an unworked shift before it starts.
A signed staying report saves your chosen reason and when you made it and we accepted it. Office staff with access to the site can read the reason, and we log that access. It can also appear in the firm's records download. It does not prove you stayed. Need help requires a connection. It can page office staff, not emergency services. Clocking out or answering a welfare check does not close the help alert.
An alert can be raised for a help request on your own eligible shift even if duty is unconfirmed. The office can see who asked and when. The alert does not prove anyone has responded.
The shift check record
A shift check records what started it, the notifications and calls attempted, your answer, its outcome and who closed it. Running late and Can't make it are recorded too. The check stores choices, numbers and times, such as an estimate of 20 minutes, not a verbatim answer. If a call transcript was saved, an office user reading the check can also see the most recent saved line from your side of the calls for that check before the answer. It is not necessarily the exact answer you gave.
WHO CAN SEE YOUR DATA explains what your office sees on the screen and in its download. Closed checks and travel estimates are due for cleanup after three years, unless a legal hold applies. A check linked to a call still on file waits until that call is removed.
MESSAGES
Message threads may be with your office, with the firm's leadership, or with people who work at a site or on a shift. The thread's membership and office access rules decide who can open it. In a leadership thread you can exclude a named owner or admin if another owner or admin can still read it. A client report link or sign-in to the client page does not open your message threads. A program your employer connects can receive message text. Your employer can pass on copies the program receives, including to a client.
You can send text or a voice note in a message. Message text, automatic transcripts and voice attachments are encrypted on our servers. A message's own sender can see app-reported delivery and read information. Office and leadership threads name readers. Site and shift threads show counts, not names. This is not a receipt from the phone's notification service.
Your employer sets how long GuardHelm keeps messages, from 30 days to 7 years. The default is one year. After that, routine cleanup deletes messages and their attachments and transcripts unless a legal hold covers them. This does not delete copies already shared.
NOTIFICATIONS
To send an alert, GuardHelm registers an address for your phone with Expo, Apple or Google. A notification can identify a shift change, announcement, time correction or licence reminder. It can include references for the recipient and work record. A new shift can say today or tomorrow and a request to stay longer can show an end time. A chat notification does not carry the message text or sender. People nearby may see a notification on your lock screen.
Where handover notices are available, a notice or reminder can go to the outgoing guard. A handover notice can say that your relief clocked in and you have not clocked out. It carries your shift references and scheduled end, not your relief's name. Expo receives those references and the recipient reference to deliver it. The notice does not clock anyone out.
A call signal goes directly to Apple or Google. It identifies the call, kind and intended membership and carries a short-lived ring token. It does not contain the guard's or site's name or conversation. Provider acceptance is not proof your phone rang. A notification already on its way cannot be recalled by signing out.
EMAIL AND TEXT MESSAGES
GuardHelm sends email through Cloudflare from no-reply@guardhelm.com. An emailed invitation link works once, for up to seven days. Your office can also hand you a setup code. Account holders may receive sign-in and security notices. Office users can request a password reset link by email.
Office staff can receive incident and shift alerts, weekly summaries, licence reminders and export notices. Some emails include names, such as a weekly late or missed shift, a licence expiry or a security change. Client report links are emailed to confirmed recipients. Invoices can go to the billing address the firm enters. The recipient's mail service handles that mail too. Mail can be saved or forwarded. A delivery status is not proof that someone read it.
Texts by phone number are off for every firm until the mobile carriers approve our number. Push notifications, in-app messages, and a text your employer sends from its own phone are not GuardHelm text messages.
Once texting is on, your office can show you the terms of GuardHelm shift texts or read them aloud. GuardHelm can text you only after you sign on paper for your number and your office records your agreement. GuardHelm queues one text to confirm your agreement. You can get a setup code and texts about your own shifts: a shift assigned, changed or canceled, or an open shift offered to you. Each of these texts starts "GuardHelm:" and names your employer.
How many texts you get depends on your schedule. Message and data rates may apply. Reply STOP to end them or HELP for help. You can also ask your office to withdraw your agreement.
Your number and the words of each text go to Telnyx, our messaging carrier, and the mobile networks only to deliver it. We do not share mobile numbers or text message consent with third parties or affiliates for marketing or promotion.
WHAT WE KEEP ABOUT YOU
Where CSV import is available, an owner can use a spreadsheet to add people, clients, sites, posts or shifts. Even at preview, GuardHelm receives the selected field values, row numbers, list type and a fingerprint of the file. It does not receive the original file, its name, column headings or unused columns. The browser can remember the column layout, not the roster. A preview alone does not add a guard or send an invitation.
For a guard this can include a name, email, phone number, starting hourly pay, licence type and number, issuing authority, issue date and expiry. Imports can also include a client's billing contacts, address or notes, sites, post orders and shift assignments. Confirming a shift import can publish assignments and notify guards. We keep the file fingerprint and which rows were added, matched or skipped, by whom and when, with a record reference where there is one. That row history has no routine expiry and is not a copy of the cells.
- Your name, your email address and any phone number your employer adds.
- Your role in your company, whether your account is active, and the history of changes to both.
- Sign-in details, including a scrambled form of your password if you have one, two-step sign-in details, and records of setup and reset codes, including who made or used them.
- Licences and training, if your office records them: number, type, issuing authority, dates, verification, notes, training details and any card or certificate it uploads. Your office can check a public register and record what it found. Licence numbers are masked in routine views. Full numbers and card images are restricted and access is logged.
- Your availability, time-off requests, and the reason or note you or the office supply.
- Hours, shift pay rates and any payroll identifier your employer sets. Office staff who may export payroll can include it in files for their payroll provider.
- Health reports about your phone: platform and versions, battery and power settings, connection, location accuracy and permissions, whether the tracker reported and how many records are waiting to send. Reports are due for cleanup 90 days after we accept them. The newest report for each phone and reports covered by a legal hold are kept longer.
- Your sign-ins: when each session started, when it was last used and when it ended, and a scrambled form of your IP address and your phone or browser type. Someone determined could reverse that scrambling, so do not treat it as strong protection.
- Call records and any saved text of calls with the assistant. If your employer has recording on, we also keep any recordings and transcripts successfully saved for those calls and for in-app calls with your office.
- Shift check records, as described under SHIFT CHECKS, with the distances and travel times worked out before your shifts, but not the positions they came from.
- Your call and alert preferences, notification addresses for this phone, and delivery records, such as when a notification or email was prepared and the provider's answer. A provider accepting it is not proof a person saw it.
- A log of access to records, such as views of off-duty positions or downloads of call recordings. This is a record of actions on GuardHelm, not of what anyone did with a copy afterwards.
- Answers to welfare checks, acknowledgements of office announcements and answers to the in-app assistant's permission question. Publishing a new privacy notice does not change an earlier acknowledgement. This notice does not ask you to agree or acknowledge it.
- Your employer's owners can see your name on the firm's GuardHelm plan statements, with the billed month, amount and any Voice charge. Those records are not your payslip.
The guard account form does not ask for a date of birth, home address, Social Security number, bank details or profile photo. Someone could still include these in a document, photo, message or other record they supply. GuardHelm uses its own device identifiers rather than the phone's serial number, IMEI or advertising ID.
WHO CAN SEE YOUR DATA
You
You can see your own receipts, hours, available pay estimates and the reports the app shows you. It does not give you the office's location map, call recordings or call transcripts. To ask for your records, see YOUR CHOICES AND REQUESTS.
Your employer's office
- Supervisors, dispatchers, admins and owners can see your punches and their positions, patrol scans, reports with photos and voice notes, hours and location within their access. Staff who cover every site can see your pay rate. Staff limited to certain sites can see hours there but cannot see another person's pay.
- They can read and search available transcripts of your calls with the assistant, whether or not your employer records calls. If it does, they can also play available recordings and read and search any saved transcripts. Their name is logged each time they open a recording or a transcript, and for each transcript a search finds. Other office staff who can open the call can see the most recent of those entries.
- Staff with the required access can download timesheets and payroll files containing names, email addresses, hours, rates, pay and payroll identifiers. Payroll exports for the whole firm require an owner or an admin covering every site. Office staff may also download shift or incident reports, call transcripts and recordings they can open. Your employer may give downloaded files to a client, an insurer, the police or a court. A downloaded copy is outside GuardHelm's control.
- Where provider pay review is available, the office can see differences for each person between GuardHelm's pay records and the amount the file would produce at the saved rates. The review can include your name, payroll ID, hours, rates and amounts. Unknown amounts are marked, not treated as zero. The office can download the review. GuardHelm does not send it to the payroll provider.
- GuardHelm uses your work hours and calculated pay, including overtime and premiums, to compare guard pay costs with what your employer bills or could bill clients. Owners and admins covering every site can see weekly totals for the firm, its clients and sites, not a list by guard. A site total can still reveal a guard's pay for the work if only one guard worked there. The weekly email can include margin percentages and the name of the site with the lowest calculated margin.
- An owner or admin covering every site can request a ZIP of the firm's records. It includes signed records and receipts, readable report text, report media, available call transcripts, permitted location records still in the live database and messages and attachments in threads that person could open when they were included. It can also include people, pay, licence and invitation records created by import. The firm's plan billing files and, where CSV import is available, the import row history and office invitation draft file are reserved for an owner's ZIP. The import files are not copies of the original spreadsheet. Call audio, archived trails and sign-in credentials are not included. The requester can download a ready ZIP for up to seven days while still authorized. Later changes to thread access do not remove messages already in that ZIP. Requests and downloads are logged; someone who saves a copy can keep it longer.
- On the shift check screen, office staff can see your answer but not the position, distance or travel time used for the check. The firm's records download can include the check reference, distance, any estimated travel time, travel mode, age and accuracy of the reading, and what the check decided. The estimate itself has no coordinates.
- Office accounts limited to certain sites can open a shift at a site they cover. If it moved there, its history can show earlier assignments, times and site or post references from another site. Your office message thread is not split by site: messages within a reader's time window can discuss other sites. For office readers, calls without a site are restricted to owners and staff who cover every site.
Outside programs connected by your employer
There are three ways to connect an outside program to an employer's schedule or office work. The sign-in connection is for the schedule and, if separately approved, for cover. A token for the office works differently, and so does an assistant connected by one person. None is the assistant that answers guard calls.
For a sign-in connection, an owner chooses the program, its sites and its permissions. The approval screen says how long the connection lasts. For a connection to Claude, Anthropic receives the results. For ChatGPT, OpenAI receives them. Connecting one does not send results to the other. If you are scheduled at a selected site, this can disclose your name and work schedule even if you never use that assistant yourself.
The schedule-reading permission lets the program read the selected sites' names, addresses and time zones, post and guard names, scheduled hours, shift status and any recorded absence for the guard and shift. It receives stable references for the site, post, guard, shift and shift version. This does not tell it who reported an absence or why. A schedule is not proof of attendance. This permission cannot change a shift or send a message.
Where a separate permission to manage cover is offered and your employer grants it, the program may advertise an unfilled shift at an authorized site and post for specified start and end times. The service can receive site, post and offer references, hours, offer or claim status, and a count of guards affected. Where its tools allow, it may read permitted openings or withdraw one it posted itself. Posting can notify eligible guards. A queued notice is not proof it was read or the shift was filled. The offer uses the firm's existing pay rules. This permission does not let the program set pay or reassign an existing staffed shift.
For a sign-in connection, a new permission needs separate authorization. A read-only grant does not gain the right to make changes when GuardHelm adds a feature. These permissions do not include fields for guard GPS, phone status, pay or personal contact details, or access to message text or incident reports. A name or site address someone enters can still contain personal information.
The approval screen states how long the connection lasts. A fixed-term grant still ends on its stated date. A continuing connection can last until the firm disconnects it or its authority ends. Ending access stops new requests. It does not undo an offer, recall a notification, or erase what the service already received. The provider's rules and the firm's account settings govern its copies, not GuardHelm's call retention periods. Ask your employer what it connected and authorized.
A token for the office
When tokens are available to your firm, an owner or admin can give a secret called a token to the firm's own software or AI assistant. The program and any service the firm sends the results to can receive them, even if you never use that program. The firm chooses a supervisor or dispatcher role and every site or particular sites. An admin who covers only some sites can choose only from those sites. GuardHelm does not ask each guard for permission or send you a notice when a token is made.
The program can read the schedule: who is assigned, at which site and post, their hours, shift status and whether an absence was recorded. It can see draft assignments before they are published. It can read open shifts, their posting and withdrawal reasons, when an offer was made or withdrawn, who took it and when, and who now holds the shift.
It can read site addresses, time zones, post and checkpoint names and which are retired. If it already has a client reference, it can see which of its permitted sites are filed under that client. With access to every site it can read people's names, roles and membership status, including people who left.
It can see items in the office's Operations list, who and which shift or site they concern, their times and whether someone needs to act. This can tell it, for example, that you were late, on a break, missed a welfare check or filed a report. A help item can mean you asked for help or that a call raised a possible safety concern. It can see when the list was checked and which sources answered. Operations can also show that a schedule change could not reach a guard or office worker because no phone was linked to them. The Operations list does not give it the GPS points, report bodies, pay fields or phone health records behind those items.
It can read office, site and shift messages within its access, including older messages. It can see a thread's title and status, who wrote each message, its kind and time, which are hidden or erased, whether the thread is unread or the office owes a reply and who is handling it and until when. It cannot read private threads with leadership or direct threads, or open message attachments.
A token limited to certain sites can read earlier site and shift threads there. A qualifying shift at one of those sites can also give it access to your office thread. That thread is not split by site: the words it sees may concern work you did elsewhere. Its office history starts no earlier than its coverage and goes back at most 30 days.
It can read announcements, who sent them and to whom, their words, urgency, where and when they apply, who is due or overdue to confirm, when someone was reached or opened the notice, and their confirmation or delivery status. An unreachable notice can tell the program that GuardHelm has no linked phone for you.
Records have names, reference numbers, times and status labels. Lists can include counts, message order, whether more records are available and where to continue. The program does not get the directory's phone or email fields, but words people enter in a name, site address, message or announcement can include personal information.
Making this token also lets its holder advertise an open shift, withdraw one it posted and start or send a message in a guard's office thread as the firm's assistant. Posting an offer or sending a message can notify people. It cannot set pay, assign a staffed shift or stop an alert reaching a person.
Each token can make up to 300 new changes in any 24 hours. Each new offer, withdrawal, office thread or message counts once. Offer and message responses can say whether the change was already made. Opening a thread on its own returns the thread, whether new or existing. A repeat that makes no new change does not count again. GuardHelm logs requests made with an authenticated token, including record references in the path it asks for.
An owner or admin can revoke a token. An admin limited to sites can revoke only tokens confined to sites they cover. A token lasts 30, 90 or 365 days unless revoked or its maker loses the authority to give it. Ending access stops new requests, not copies the program or its provider already has, or notices or messages already sent. Ask your employer which program holds it and what it does with its copies.
Assistants connected by one person
When your employer allows it, a person who works in the office can connect Claude or ChatGPT to their own GuardHelm account. Each person decides for their own account. The assistant can read and do only what that person can read and do in GuardHelm, and only for as long as they keep it connected. It is not the assistant that answers guard calls.
The assistant can read the office records that person can see. Depending on their role and sites, that can include names and contact details, schedules and attendance, the text of messages and announcements, incident and other report text, licences and the details of documents on file, hours and pay, clients and invoices, and account activity. If you work for the firm, this can include your messages, your reports and your pay even if you never use the assistant yourself.
Everything the assistant reads goes to the company that runs it: Anthropic for Claude, OpenAI for ChatGPT. Connecting one does not send anything to the other. If the firm lets a person use a personal token instead, what it reads goes to whichever program that person gives the token to. Their terms and the firm's account with them govern what they keep, not GuardHelm's retention periods. A file opens only in GuardHelm for the person to download; the assistant does not receive its bytes.
Precise work location is separate. It starts off, and the connecting person must choose to share it. Worker privacy choices and site restrictions still apply.
Changes come in two kinds. On an owner's own connection, routine changes happen straight from the chat, such as adding people, sites and clients, scheduling and ordinary edits. An owner can turn this off for each connection. Every change made this way is listed in GuardHelm with who made it and what changed, and where a change can be reversed it has an Undo button there.
Money, pay, texts to guards, a report sent back to its guard, removing, suspending or changing a person's access, anything sent to or shared with someone outside the firm, and the firm's settings for closing, exports, record keeping, call recording and tracking always wait until the person confirms the exact change in GuardHelm. So does every change on a connection that is not an owner's, and any change that needs a fresh sign-in check.
For six hours after the assistant reads text people wrote in GuardHelm, such as a message, report or document, every change it asks for waits until the person confirms it in GuardHelm. To tell, GuardHelm records when such text was given to the connection, not the text itself, for six hours.
GuardHelm records each connection, what it read and changed and the outcome. An unconfirmed change expires after 24 hours. Disconnecting stops future access. It does not undo a change already made or erase what the assistant already received. Ask your employer which assistants are connected and who receives the results.
Full control for an owner's assistant
An owner can turn on full control for their firm. While it is on, an assistant that an owner connected to their own account carries out every change it asks for at once, without the owner confirming it in GuardHelm. That includes money, pay, texts to guards, anything sent to or shared with someone outside the firm, changing or removing a person's access or role, and changes made after it read text people wrote in GuardHelm. The waits described in the section above, including the six hours after reading text, do not apply to those connections while full control is on.
Even with full control on, the assistant cannot change an owner's password, authenticator or device keys, remove or demote an owner, make someone an owner, close the firm's account, export all of the firm's data, or place or release a legal hold. It also never makes a legal statement in the owner's name, such as confirming that guards agreed in writing to location tracking.
Every change made under full control is recorded with who made it, what changed and when, and where a change can be reversed it has an Undo button in GuardHelm. The firm's owners are emailed a list of these changes each day. Any owner can turn full control off at any time. Ask your employer whether it is on.
Your employer's clients
Your employer can have GuardHelm email report links to a confirmed client address. Anyone holding such a link can open it without signing in for up to 45 days, unless access ends sooner. It can also open other still-released reports for that client at that site, up to 90 days back. GuardHelm logs eligible report reads, the address tied to the link and the time. We log eligible photo requests by work record, not image. An entry does not prove the file was served. A forwarded link does not tell us who used it.
A report can show coverage, the post and times, patrol activity and incidents. A released incident may include the office's current version of the text unless the office leaves it out. By default clients see photo counts, not the photos. The office can choose photos for a client. A shared site log shows answers and photo counts, not the log photos.
An owner or admin who covers every site chooses what to share with each client. By default a report does not name you as the guard, and its location detail is the site only. The office can show your name. A daily report can show checkpoint names and scan times. If the office chose Checkpoints or Path before precise locations were available, that choice does not share exact checkpoint locations or your phone's path. Making the options available or saving other settings does not add either to a report.
To share exact checkpoint locations where known, the office must make a new choice for that client. It can share those locations alone or choose Path to share them along with your path. If the office turns that detail off, sharing it again takes a new choice.
With Path, the browser receives eligible phone positions and their times on the client's site while you are on the clock and not on a recorded break. The report draws the route without displaying the time beside each point. The fence check allows for the accuracy of the reading near the edge. If there is no fence, no path is shared. The line joins available points, not every step, and does not join across parts of the shift left out.
The office's choices apply the next time someone opens a report, on the client's page or from a link that still works, even one sent before the choice. They do not change a saved copy. You cannot set them in the guard app. Ask your office what it shares. Reports do not give a client the live guard map or your pay. A name or other detail may still appear in words or photos a person shares.
A client can use the address on file for reports to ask for a separate email to open its page. The address must be confirmed at a site where a report send within the past 45 days still qualifies. A failed delivery can count. The person does not need to have received the report link. Asking for a sign-in email does not prove one was sent.
The newest sign-in link works once, for 15 minutes. Someone given that link can use it too. A browser cookie then lasts up to 12 hours. The page can show other released reports and shared photos for sites where that address was confirmed before sign-in and still qualifies. It can also show a client's invoice or credit note if the billing address matches, its email send was accepted and that invoice link still opens.
Your employer can choose to show posts and shift hours for shifts under way or starting in the next 12 hours. This is off by default. The coverage status can use punches, an office confirmation, the schedule or a call-off. It does not establish who is on duty now. No guard name or position is shown in that list.
We log successful sign-ins and eligible page reads with the address, time and site or record where applicable. Repeat opens can count as one. Signing out clears the browser's cookie and ends the session if the address still qualifies. If the address stops reports or your office removes it at a site, it loses access there. Without an eligible site the page will not show the firm's records.
A receipt link is separate. Anyone holding it can open it for up to 45 days unless access ends sooner. The copy can show the firm, client, site or post, type of work, times and verification details. The office's filing site can differ from the site in the signed record. Where the copy distinguishes them, the original site or post may be withheld. It may say whether a punch or scan was at the site. The receipt has no field for your name, exact coordinates, pay or report text.
A holder can save the signed receipt and any available independent timestamp proof. The built-in browser check verifies GuardHelm's countersignature, not the phone signature or the independent timestamp. Expiry or revocation stops new opens, not copies already saved. Your employer can also give a client a file it downloaded, such as a call recording.
A client bill for hours can show the site and post, work times, billed hours and rates, an approval time, and evidence from clock-ins, clock-outs and breaks. These are billing rates, not your wages. Where your employer bills by verified patrol visit, GuardHelm uses duty records and signed checkpoint scans to check which rounds were due and completed. It keeps the evidence with the issued invoice. The client copy can show the billed site, when the round was due, the charge or credit, scan times and receipt fingerprints. A scan does not show how long you stayed.
The invoice and its data file are separate from client reports. The bill can include the billing contact's name, email and address, the client's name and code, and the firm's contact details, tax ID and payment instructions. The data file can include signed receipts with firm and work record references, when the phone recorded the punch or scan and when GuardHelm accepted it. It can show a punch's coarse position relative to a site, a scan method and verification results.
Anyone holding an invoice link can open the bill or its data file without signing in for up to 180 days unless the office revokes it sooner. Report sharing settings do not change that copy. For a public invoice link we log the link, time and file type, not who used it. On the signed-in page we also log the address and invoice. Ending the link does not remove the invoice or recall a saved copy. Where credit review is available, an hours or visit credit that exceeds the amount or tax left to credit on an invoice is held for office review. It is not issued in part or shown to the client as a credit on the bill.
Where readiness packets are available, an owner or admin covering every site can prepare a dated packet for a client on file and give it to an authorized client or prospective client. The roster starts with counts, not names. The hosted service in this release does not offer named rosters. Where the figures are available, a packet can show seven days of scheduled guard-hours and covered post-hours. Overlapping shifts at one post count once for covered post-hours. These are schedule figures, not attendance or pay.
If named rosters become available to your firm, the office could list you by name under a post whose recorded requirements you met when it previewed the packet. This is not a fresh check of your credentials at download. It does not mean you worked there, are available or have agreed to take the post. The roster would show the required licence or training type and any recorded expiry, using the earlier date if the office has both a credential expiry and a registry expiry. It could also show the time and result of the last registry check the office recorded for that credential. GuardHelm does not check the register itself or certify your eligibility.
Your licence number would appear in the roster only if the office separately chose to include numbers in that packet. The office can also attach full copies of selected insurance certificates from the client's file. These are included as filed, without redaction, even with a counts-only roster. A name or number in a certificate can therefore be shared.
GuardHelm logs who previews a packet, when, for which client, what they selected and which guards the roster names, even without a download. If licence numbers are shown, the log identifies the credentials, not the numbers themselves. Preparing the PDF is logged separately before the browser saves it. Your employer decides who gets the file. GuardHelm does not send it to a client, and its log does not say who received it.
Other security companies
Having an account at another security company does not give someone access to your employer's records. Access in GuardHelm is limited by company and role. Files and links your employer chooses to share are separate.
GuardHelm
People at GuardHelm with production access can reach what is stored. GuardHelm keeps the master encryption keys separately. Encrypted record keys can be stored alongside the protected data. A firm's records download can include a readable signed record and the key used to check its contents. Recent tracking points in the database and points waiting for processing are not separately encrypted by GuardHelm. Older points in the daily trail archive are encrypted.
During an assistant call the working text is also held in Cloudflare storage without GuardHelm's own encryption. GuardHelm clears it when it finishes closing the call and saving any transcript. If either step fails, it retries that step for about a day before giving up. Cloudflare recovery copies can persist for up to 30 days. Office exports contain readable material and anyone holding a downloaded copy can read it.
SUPPLIERS
These companies help us run GuardHelm and handle your data as part of doing so:
- Cloudflare: servers and network, file storage, older location archives, email delivery, in-app calls, AI for voice notes, report write-ups and automatic checks, the call notice and the assistant's backup service, and server logs.
- Telnyx: our messaging carrier. Once texting is on, it and the mobile networks get your number and the words of each text to deliver it. A confirmation is not queued again for the same agreement or while that number has stopped texts. A reply to a HELP text you send is separate from a text your office initiates.
- OpenAI: AI models that hear and answer calls with the assistant.
- Google: AI models that hear and answer calls with the assistant, run on Google Cloud in Iowa, USA.
- Neon: the database, hosted on Amazon Web Services in Ohio, USA.
- Expo, Apple and Google: notifications and call signals to phones. Apple and Google provide phone location services. If you tap Get directions, the maps service opened by your phone receives the site's address.
- Google, Mozilla, Apple and Microsoft: deliver console alerts to office staff's browsers when enabled.
- Transistorsoft: makes the location component built into the app. The tracker sends its points to GuardHelm. We have not confirmed whether the component contacts its maker, for example to check its licence.
- For satellite and aerial imagery, see OFFICE STAFF USING THE CONSOLE. For programs connected by your employer, including Claude or ChatGPT, see WHO CAN SEE YOUR DATA.
- Stripe: billing details when we enrol a firm, and card payment of the firm's own GuardHelm bill. Enrolment can come before card payment is switched on. See OFFICE STAFF USING THE CONSOLE.
- The office can download a billing CSV for QuickBooks with client, site or post, service dates, amounts and, where available, visit counts and prices. A download does not send it to Intuit. If the firm separately sends a bill or credit to its connected QuickBooks Online account, Intuit can receive client and invoice references, service dates, billed rates and hours, tax and the office's credit description. Intuit can also receive a purchase order number and the billed site and post for hourly lines. Visit lines can give the billed site, due time, time zone, quantity and charge or credit. The punch and scan receipts do not go with that send. Downloading a payroll file does not send it to a payroll provider either.
We do not sell your data. GuardHelm uses no third-party advertising or analytics code in the guard app or office console. Email and delivery by text are described above. Texts stay off until the mobile carriers approve our number.
A crash in the guard app or an error in the console can send GuardHelm the app, build, screen name, kind of error and up to five code locations. The report has no field for the error message, work records or personal names. The network address is used to limit repeated reports. The report goes to GuardHelm's server logs at Cloudflare, not to a separate crash-reporting service.
WHERE YOUR DATA IS HELD
The database is in the United States, in Ohio, and our servers are set to run beside it. Some work, such as speech-to-text and parts of calls, runs wherever Cloudflare's network handles it. Your phone's requests first reach the nearest Cloudflare location, which can be outside the United States. Cloudflare stores photos, voice notes, call recordings and other files, and we have not chosen a region for them.
Google handles calls with the assistant in the United States, in Iowa. We have not chosen a region with OpenAI, so it may handle them in other countries.
ON YOUR PHONE
- To work without signal, the app keeps signed records and receipts, shift and schedule details, post orders and notes, site contacts, patrols, hours, messages and reports in progress on your phone. A queued record is not yet a record received by the office.
- The app encrypts report text, unfinished reports and punch positions stored on the phone. Its key is kept on that phone and is not included in phone backups. This is separate from copies a person shares outside the app.
- Once the office accepts a signed record, the app can remove its local body and key but keep the receipt. An ordinary completed sign-out clears the guard's local work data. The app can refuse to sign out while you are clocked in, work is waiting, a report is unfinished, or tracker cleanup is unresolved. Ending an unverified replacement sign-in can leave the original guard's work protected on that phone. Signing out does not erase copies you saved outside the app.
- Do not delete the app to clear a queue. Ask the office for help when work is still on the phone.
- Tell your office if you lose a phone. You can also report your own missing phone from another eligible phone. Marking it lost retires its keys and tracking credentials and can sign you out on your other devices. An otherwise eligible punch signed with a previously active key can still be kept as a flagged exception after the phone is reported lost. Other late signed records can be set apart for review. GuardHelm cannot supply a copy of work that never reached it.
HOW LONG WE KEEP IT
- Background location points: about 90 days.
- Recent points for the live map: normally about five days. This is not the deletion period for the underlying location history.
- Phone reports: 90 days, except the newest one. That is when routine cleanup can begin after acceptance, not a deletion deadline. A legal hold can keep a covered report longer.
- Call records, call recordings and call transcripts, including assistant transcripts: one year. Cleanup is not immediate and a legal hold can keep them longer.
- Call data at OpenAI or Google: see CALLS for what the provider may keep and what GuardHelm can ask it to delete.
- Shift check records, with the distances and travel times in them: 3 years. Closed checks age from closure, estimates from when they were made.
- Messages: what your employer sets, from 30 days to 7 years. The default is one year.
- Where CSV import is available, import row history has no routine expiry. Unsent office invitation drafts expire after 30 days; their details can remain until cleanup removes them. Issued drafts keep a reference to the invitation, not the proposed person's details.
- Work records, the photos and voice notes in reports, receipts and timesheet approvals: no end date set.
- Your account, your sign-in history, the log of who saw your data, and your acknowledgements: no end date set.
- Automatic check answers: 90 days while a check is being tested, otherwise as long as the record they are about.
- A legal hold keeps covered records beyond their usual cleanup period. It does not keep every copy, draft or download available. The live map can stop showing a point while the underlying evidence is kept.
When something is deleted, copies can stay in our suppliers' backups for a limited time until those backups expire.
IF YOU LEAVE YOUR EMPLOYER
When your employer removes you, you lose office access and GuardHelm stops accepting new background tracking for you. An offline phone might not know yet. Your work already on file stays with your employer. Ending access does not necessarily erase the copy still on your phone or prevent previously saved, signed work from being submitted. Ask the office before clearing or giving away that phone.
A whole company closing is different. Its owner can name a last day and cancel the closure before then. After that day, owners and admins have 30 days of read and download access. Saved work may still arrive from phones. The owner should obtain and check the final export; starting an export does not establish that anyone received all of its files.
Closing does not itself erase the firm's records. After a written request and checks on the final export or the owner's waiver, a separate GuardHelm procedure can remove the closed firm's records and queue its stored files for deletion. While any legal hold is active for the firm, that procedure refuses to erase the firm or queue its files. There is no override. There is no automatic erasure date, and this is not a way to delete a single person's shift. An erasure log and an identity still used by another firm can remain. Receipt timestamp fingerprints, batch details and replies are kept separately, as described under WORK RECORDS.
YOUR CHOICES AND REQUESTS
- Location: your phone's Settings, or ask your employer.
- The assistant: say yes or no when the app asks, and change your answer on the Help screen. Withdrawing permission can stop the connection even if a ring already reached you.
- Shift checks: your employer chooses whether to use them and whether to use available positions before shifts. You can set call preferences. See SHIFT CHECKS for the calls that can come during quiet hours.
- Call recording: your employer decides, and there is no switch for you. GuardHelm never records a call you make with your phone's own dialer.
- A phone copy: Send myself a copy on Receipts or Hours shares a list of your records or hours. If access has ended, any phone copy contains only what was saved on that phone. This is not the office's personal records download.
- A fuller copy: ask your employer. An owner or admin who covers every site can record your request and download a ZIP of records linked to you. The request and download are logged separately. The office reviews and gives you the file. Recording a request is not proof a file reached you.
- The ZIP can include your shifts, punches, pay rates, licences, reports, messages you sent and permitted location records. It is not a copy of everything. It excludes original signed files, media attachments and call transcripts. If GuardHelm cannot safely tell whose records they are, the office must check the originals rather than send a guessed match.
- Leaving the firm: in the app, open Help, tap App and account, then Leave this company. Leaving is not deletion of your work records.
- Anything else: write to hello@guardhelm.com with your name and your employer's name to ask for a copy, correction or deletion, or to ask about this notice. A request is not a promise of deletion. Work records stay with your employer and it may need to keep other records too.
OFFICE STAFF USING THE CONSOLE
We keep your name, email address, role and the sites you cover, your sign-in details including two-step sign-in, your sessions as described above, and a log of what you open and change. Office users cannot edit or delete that log. It can be removed as part of whole-firm erasure. Calls you make or take in the console are kept as call records.
Where CSV import is available, an owner can create an office invitation draft before you have an account in the firm. The draft can contain your name, email, proposed role and sites or whole-firm access. It does not grant access or send an email. An unsent draft expires after 30 days and cannot then be sent. An expired draft can appear in a new owner export until cleanup removes it. If the owner issues an invitation, GuardHelm removes those details from the draft and keeps a reference to the invitation. The invitation itself still contains the details. A ZIP already made can still contain details that were later removed. Issuing it does not prove the email was delivered.
The guard's background tracker does not run on office accounts. If you are on a supervisor watch, the phone app can send your position while the watch screen is open, or the site you select instead. Your office can see the site and time. A post check can also include your phone's position. Watch positions and post check coordinates are cleared by a nightly job after 90 days unless under legal hold. The check itself stays.
Satellite view can load standard imagery from the US Geological Survey, which sees your IP address and the area you are viewing. GuardHelm serves sharper aerial photos itself where they are available.
The office may keep client billing contacts, invoice and credit note details, amounts, payments and records of invoice delivery and link opens. WHO CAN SEE YOUR DATA explains the work evidence on client invoices, including what a link holder can download. Client billing is separate from guard pay. A firm can give payroll files to its chosen payroll provider. GuardHelm does not run payroll in this installation.
When we enrol a firm for card payment of its own GuardHelm bill, we send Stripe the firm's billing name, email, business address and firm reference. This can happen before card payment is switched on. It is not a charge to guards or a payment of wages.
Card payment is off by default. When it is on for an enrolled firm, an owner must pass a recent identity check to open Stripe's page. There the owner can see past bills, manage a card, change the billing email or supply a tax ID. Stripe handles the card details. GuardHelm does not keep the full card number or security code. It can show the owner the card brand, last four digits and expiry that Stripe supplies. If the default is a bank account, it can show its last four digits. GuardHelm does not debit that account.
We start automatic collection only for eligible bills for closed months, from the firm's agreed first billing month, with a default card. Stripe receives the firm, customer and statement references, the billing month, counts of guards at each plan rate and the amounts billed, not guard names or individual work records.
At the first collection decision, an otherwise eligible bill goes to manual payment if Stripe confirms there is no default card. If the card is gone at the final check, we hold the bill. A bill sent for manual payment is not charged automatically later. We do not start automatic card collection for a month with a suspension or one that ended while the firm was closed. Turning off new collection does not stop Stripe from retrying a bill it is already collecting.
GuardHelm can email a bill for manual payment to eligible owners through Cloudflare. It sends a separate statement summary to GuardHelm's billing contact. If a card payment fails or the bank needs confirmation, GuardHelm can email eligible owners the amount, reason and next attempt if known. A request to confirm is not a completed charge.
We keep the firm's enrolment, Stripe customer and invoice references, amounts, tax, remaining balance, payment status and any failure, due date or next attempt. We also keep Stripe event references and types, a log of requests to open the payment page and records of billing email delivery attempts.
To get a sharper photo, GuardHelm can ask the US Department of Agriculture for a map rectangle around a work site. The service sees the approximate site location and the time of the request, which can follow a fence or checkpoint edit or a scheduled update. It may see that GuardHelm made the request. We do not send the firm or site name, street address, fence outline, or any guard's identity or phone location in that request.
On a recorded office call, your side can also be recorded. If a transcript is saved, your side is labelled with your name. The console plays you the same recording notice as the guard. Recording starts with it, and the console shows "Recording". People near your computer can be recorded even if they did not hear the notice. Tell them the call is recorded. The same access, logging and retention rules apply as for the guard. Staff who can open the call can see the most recent entries for who opened or downloaded its recording or transcript.
If you turn on browser alerts, the service run by Google, Mozilla, Apple or Microsoft for your browser delivers them. An incident alert can name a site on the lock screen. The payload is encrypted for your browser. Other alerts can carry a date, an event type, or no details. If you enable incident email, a message can also go to your office email address.
Office phone alerts are on by default for owners and can also go to staff on watch even when their switch is off. Other eligible office staff can turn on phone alerts. The lock-screen text is generic. References in the notification help open the right alert or incident. GuardHelm records your alert preference and the attempt to notify you.
CHANGES TO THIS NOTICE
This version adds full control, which an owner can turn on so their own assistant acts without asking. Version 2026-11-21d added automatic checks on voice notes to your office, report drafts and post orders. Version 2026-11-21c added automatic checks on incident reports, assistant calls and photo stops. Version 2026-11-21b added assistants that one office person connects to their own account, and said GuardHelm itself sends every text, naming your employer inside, once texting is on. Version 2026-11-21a added the off-duty trail an office can see while tracking is set to Always. Earlier versions explained access by outside programs, receipt timestamps, client reports and bills, readiness packets, spreadsheet imports and payment of the firm's GuardHelm bill, what the office can download, what is recorded during calls and shift checks, and how legal holds affect cleanup and erasure.
Each version is stored unchanged, with a fingerprint of its exact text. A change means a new version. Publishing leaves earlier versions and acknowledgements unchanged. It does not record anyone's consent or ask for a new acknowledgement.
CONTACT
WASAW LLC, doing business as GuardHelm
418 Broadway STE N, Albany, NY 12207
hello@guardhelm.com