Data Processing Agreement
Last updated 16 November 2026
This agreement covers how Peoplume processes personal data on your behalf. It forms part of the terms of service and takes precedence over them, and over the privacy page, on any conflict about processing.
It binds both of us as soon as you accept the terms — there is nothing to sign. Sign-up shows the terms and this agreement beside the button that creates the account, and records which version (the date at the top of this page) was accepted, when and by whom; the Billing page shows it. If your procurement needs an executed copy, write to [email protected] and we will sign this text as it stands.
1. Who is who
You are the controller of the personal data in Annex I. You decide who is recorded, what is recorded about them, what documents are uploaded and how long any of it is kept. Plumeware LLC, of Wyoming, United States, operating Peoplume, is the processor and acts only on your instructions.
For our own account and billing data — who your administrators are, your subscription status and seat count — we are an independent controller and this agreement does not apply. That processing is described on the privacy page.
You confirm you have a lawful basis for what you record, and that you have given your staff whatever notice your local law requires. We cannot establish that basis for you and do not try to.
“GDPR” means Regulation (EU) 2016/679 and, where it applies, the UK GDPR and Data Protection Act 2018. “Personal data”, “processing”, “controller”, “processor”, “data subject” and “personal data breach” carry their GDPR meanings.
2. Our instructions are yours
We process personal data only on your documented instructions, including about transfers to another country, unless a law we are subject to requires otherwise — in which case we will tell you before processing, unless that law forbids it on important grounds of public interest.
Your instructions are: the terms, this agreement, whatever you configure in the product, and any further written instruction we accept. If we think an instruction breaches data protection law we will say so, and may pause it until it is withdrawn or changed.
We do not use your data to train AI models, and we do not sell it. A model sees your data in two optional places, both in Annex III: the receipt-reading feature, which is off until an administrator switches it on and then runs only when someone presses the button for one receipt (in the portal or the phone app alike), and the AI assistant, which is off until an administrator switches it on and then sends each question with the rows looked up to answer it — never more than the asking person could already open. Anthropic processes that data under its own terms, not ones we negotiated, and we do not promise that it will not use it to train or improve its models (Annex III); an administrator who does not accept that leaves both features off.
3. Confidentiality and access
Everyone authorised to process personal data under this agreement is bound by confidentiality, and access to production data is limited to the people who operate the service and only where they need it.
4. Security
We implement the measures in Annex II, taking account of the state of the art, cost, and the nature and risk of the processing, as Article 32 requires. We may change them, but not in a way that lowers the overall level of security. Annex II also names what we do not do — a security annex listing only strengths tells you nothing.
5. Sub-processors
You give general authorisation for the sub-processors in Annex III. Each is bound by terms no less protective than this agreement, except Anthropic, which handles data for the two optional AI features under its own commercial terms (Annex III), and we stay fully liable to you for what they do.
We will give you thirty days’ notice, by email to your administrators, before adding or replacing one. If you object on reasonable data protection grounds within that window we will look for an alternative; if there is none, you may stop using the affected part of the service and we will refund the unused part of what you have prepaid for it.
6. Helping you meet your own obligations
Requests from individuals. The product answers most of these without us. You can export one person’s records as a set of spreadsheets from their profile (the privacy page lists what it holds and what it leaves out), correct any field, and erase a former employee’s personal data while keeping the payroll record. Employees can download their own bundle from the portal without asking anyone; it holds back, and names as held and not given, the 360° feedback colleagues wrote about them (written for whoever writes their review, never shown to its subject), succession plan entries naming them (your planning record), suggestions of colleagues to give peer feedback on their review or naming them as one, with what HR decided (who is asked to give feedback is not shown to the person it is about), the private notes their manager wrote on their one-to-ones (never shown to them), the note a manager wrote for HR with a recommendation on their probation, the assessment from their job applications (interviewers’ feedback, ratings and scorecards, the hiring manager’s recommendation and the recruiter’s and interviewers’ notes, with an offer’s terms until it is sent to them (and again from a change to a sent offer until it is sent again) and every change of its terms and its approval steps, written for the people deciding on a hire), and job openings and requisitions naming them as the person a new hire replaces (your hiring plan) — so whether to disclose those in answer to a request is your decision, made with the export you download from their profile. In that export each sensitive part needs the permission of the page that shows it: self-assessments and 360° feedback need permission to write reviews, one-to-ones and suggested peer reviewers permission to manage performance, job applications and hiring permission to view recruitment. It never holds rows about the person downloading it (360° feedback about them, suggestions of peer reviewers for their own review or naming them, and interview feedback the colleague wrote, or recommendations they made, on the reader’s own job application). Someone exporting their own record from the console through an account linked to it gets the employee’s own download, with the same seven things held back; a console user whose sign-in email is merely on the record gets only what their own permissions allow, never those seven, and that export is logged as any export of someone’s record is. For a job applicant, from their application page: export everything held under their email address, or delete every application under it at once with its CVs, interviews, feedback, offers, messages and notifications — refused while one of them was hired (erasing that employee’s record covers it) or holds a sent or accepted offer. Where a request cannot be answered through those features we will assist you as far as we reasonably can.
If one of your employees contacts us directly we will not answer on the substance — we will point them to you and tell you promptly.
Breaches. We will notify you without undue delay and in any event within 48 hours of becoming aware of a personal data breach affecting data processed under this agreement. The notice will describe what happened, the categories and rough number of people and records involved, the likely consequences, what we have done or propose to do, and who to talk to. Where we cannot give you all of that at once we will follow up as we learn more. Any notification to a regulator or to the people affected is yours to make — you are the controller and only you know the context.
Impact assessments. We will give you the information about the product you reasonably need to carry out a data protection impact assessment, and to consult a supervisory authority where Article 36 requires it.
7. Deletion and return
You can export your data at any time, including after cancelling, and can erase an individual or delete records outright. Those are self-service and take effect immediately.
Otherwise we keep your data for 90 days after the agreement ends so you can retrieve it, then delete or return it at your choice, unless a law requires us to keep something — in which case we will tell you what and why. Return is by export in a structured, commonly used, machine-readable format: the data archive (CSV files, with a README naming what it holds and what it does not) and the uploaded-files download (the files themselves, with a manifest saying which record each belongs to). Each holds what the permissions of the person downloading it allow, so one of your Administrators downloads all of it except the records about themselves — their own job applications and their CVs, requisitions naming them as the person a hire replaces, their places on hiring teams, one-to-ones about them, peer reviewers suggested for their own review and the note for HR on a probation recommendation about them — which are left out of their own download and named in its README. A second Administrator’s download completes the return.
Deletion on request is as in the Terms, section 9: one of your Administrators asks from the Billing page, once the subscription no longer renews and the payment provider is no longer collecting, and no sooner than seven days after the request we delete every record of the company and every file it uploaded from the live database and file storage, keeping only a record that it was done.
Of what we hold, backups and space freed inside the live database file are the exceptions to immediate deletion. Data erased from the live system stays in existing backup copies until those rotate out. Backups are used for nothing but restoring the service and, each day, checking that the newest one reads back whole (Annex II); that check keeps no copy. Within the live database file itself, the space a deleted record — or a value since changed or blanked — used can keep its old contents until that space is reused or the file is compacted, except where Peoplume zeroes it as it frees it: erasing an employee’s personal data, deleting an employee record, and rewriting a survey’s responses — and deleting a whole company. That space is not readable through the product. Beyond the live database and file storage, deleting the company or erasing a person does not reach what a sub-processor (Annex III) already received in doing its job — for example the emails already sent through Resend, and what receipt reading or the AI assistant had already sent to Anthropic. Anthropic keeps or deletes what it received under its own terms (Annex III) — its published policy is to delete it within 30 days, with the exceptions listed there — and we do not promise that it will not use it to train or improve its models, or keep what it uses for that.
One deliberate limit on erasure: erasing a former employee blanks their identifying fields but keeps payslips, attendance and salary history, and the receipts of expense claims that were reimbursed — the evidence for money that was paid (receipts of claims never paid — rejected, including an approval later withdrawn, or sent back and not resubmitted — are deleted, as are claims in a trip report the person never sent, with their receipts, and those trip reports, together with any declined or cancelled travel advance on them; erasure waits while any claim is still pending, while an expense claim or air-ticket cash-out is approved but not yet paid, while a travel advance is still requested, approved, handed over or being settled, until the day after the person’s last day, and while a console account tied to the record — linked to it, or signing in with its email — still has access, so ending that access in Settings comes first). Leave history stays too, including stated leave figures, whose days are kept but whose typed reasons are blanked, and leave requests, whose dates, days and reasons are kept but whose approvers’ and HR’s notes are blanked — and, where you have a leave type limited by gender (women only or men only, an archived one included), so does their gender, because it decides which leave they had, so their balances and settlement do not move; so do the trip reports they sent and their other travel advances, with their titles, dates and amounts, while the purpose and the notes typed on them are blanked. The job application they were hired from — the one their record was made from or, where no record is tied to it, a hired application under the email on their record, when no other record not yet erased carries that email (whether that person still works there or has left) — and any application on which they were an internal candidate lose their name, email, phone, CV, their name as the one who applied, their request to be kept for future roles, the offer’s response links, the emails written to them (the history keeps that one was sent, without its subject), the free text (recruiter, history, detail-correction, interview, feedback, offer and self-scheduling notes, the note on a hiring manager’s recommendation, and a reason typed when declining an offer online, cut back to “declined online”) and the staff notifications that link to them, and keep their stage, dates, source, outcome reason, ratings, interview times and places and offer terms; a link between such an application and one that stays named (one put forward from the other) is cut, with the opening each one’s history names for the other. Their places on hiring teams and their unsubmitted interview feedback drafts are deleted; interview feedback they submitted, recommendations they made as a hiring manager and an application’s record that they referred the candidate stay, under the erased record’s label (each recommendation’s entry in the application’s history keeps the name it was recorded with, as staff names on records they acted on do); openings and requisitions naming them as the person a hire replaces keep the record and lose their free text (why the seat was needed, a decision note). Praise written by or about the person is deleted with its notifications, their places as a successor are deleted, their place as the head of a department is cleared, the notes and flight-risk judgement on the succession plan covering their role are cleared (the plan stays), and their name and login email on reviews they submitted about colleagues are replaced by the erased record’s label, a review a cycle still holds for calibration included (the reviews stay; a review is theirs when it was submitted from a login tied to their record — their portal login, or a console account linked to the record or, linked to none, signing in with its email). Their review, self-assessment, 360° feedback, goal and training records are kept, text included, because they are evidence in employment disputes; their text may still name or describe the person. So are their comment on a review they were shown, their self-ratings and answers to a cycle’s questions, the reasons they gave asking for training or for a certificate to be renewed, and the probation decisions about them, whose typed note is blanked; the certificate files on their training records are deleted, and so is the name of whoever created each of their goals (on a goal they proposed, their own). Their one-to-ones — shared and private notes and action items — are deleted as informal working notes, as are suggestions of peer reviewers about their review or naming them; the note on a manager’s recommendation on their probation is blanked (the recommendation stays); improvement plans about them are kept with their check-ins and text, as a formal performance record; and their name and login email on one-to-ones, probation recommendations and improvement-plan check-ins they recorded about colleagues as a manager, and on peer reviewers they suggested, are replaced by the erased record’s label (the records stay; each is found by the employee record it was written from, so after their login has been removed too). The names of staff recorded on the records those staff acted on — who added someone to a review cycle, returned a review, recorded a probation decision, appointed a successor or created a colleague’s goal — are kept, as the audit log keeps the names it recorded. A right to erasure does not entitle anyone to rewrite what a company reported it paid, and deleting payroll records would change an accounting record after the fact. The audit entry recording the erasure keeps the name, so you can still answer “did you action my request?” a year later; the audit log’s other entries keep the names they recorded too, and the entry recording an administrator’s removal of a praise quotes its author and opening words.
8. Audits
We will give you the information needed to show we comply with Article 28 and this agreement, and will allow for and contribute to audits by you or an auditor you appoint.
In practice: we answer security questionnaires and provide documentation on request. A deeper audit may be carried out once in any twelve months, on thirty days’ notice, in business hours, without unreasonably disrupting the service, subject to confidentiality and at your cost — except where the audit follows a breach we notified or a regulator requires it, when none of those limits apply.
9. International transfers
We are established in the United States, and we are not certified under the EU–US Data Privacy Framework. The European Commission’s adequacy decision for the United States covers only organisations that hold that certification, so it does not cover transfers to us, and we would rather lead with that than leave you to find it. Personal data is also processed by the sub-processors in Annex III, in the regions stated there.
Where you are established in the EEA, the United Kingdom or Switzerland, or the GDPR otherwise applies to your processing, the Standard Contractual Clauses in Commission Implementing Decision (EU) 2021/914, Module Two (controller to processor), are incorporated into this agreement and apply to transfers to us, completed as follows:
- Clause 7 (docking clause) applies.
- Clause 9: Option 2, general written authorisation, with the thirty days’ notice in section 5 above.
- Clause 11: the optional independent dispute resolution paragraph does not apply.
- Clause 17: governed by the law of Ireland.
- Clause 18(b): the courts of Ireland.
- Annexes I, II and III of the Clauses are the correspondingly numbered annexes below.
For transfers under the UK GDPR, the UK International Data Transfer Addendum (version B1.0) applies to those Clauses, with Table 4 completed as “neither party”. For Switzerland, references to the GDPR are read as references to the Swiss FADP and the Federal Data Protection and Information Commissioner is the competent authority.
As Clause 14 requires, we each warrant that we have no reason to believe the laws and practices of the destination countries, applied to the transfer and to the measures in Annex II, stop us meeting our obligations. We will document that assessment, make it available on request, and tell you without delay if we become unable to comply.
10. Other data protection laws
Where another data protection law applies to this processing — a United States state privacy law, or the law of a country where you or the people you record are located — we comply with the obligations it places on us as a processor or service provider. We do not sell the personal data you put into Peoplume or use it for any purpose other than providing the service to you. Nothing here reduces a protection such a law gives.
11. Liability, term and changes
Liability under this agreement is subject to the limits in section 10 of the terms, except where the GDPR or the Standard Contractual Clauses do not permit that — the Clauses govern liability between us for the transfers they cover.
This agreement starts when you accept the terms and runs until the service ends and deletion under section 7 is complete. Sections 3, 7, 8, 9 and 11 survive. We may update it to reflect a change in law, in the product, or in our sub-processors; material changes are notified as the terms require, and we will not apply a change that reduces your protection or our obligations without your agreement.
Annex I — What is processed
Parties. Exporter and controller: the organisation that accepted the terms, as employer of the people recorded; contact is the administrators on the account. Importer and processor: Plumeware LLC, Wyoming, United States, operating Peoplume; contact [email protected].
Data subjects. Your employees and former employees, their dependents where you record them, job applicants — including people who apply through your public careers page — and the administrators and managers who use the console.
| Category | What it includes |
|---|---|
| Identity and contact | Name, employee number, work email, phone, address, date of birth, gender, nationality, marital status, emergency contact name and phone. Everything except name and employee number is optional. |
| Employment | Hire date, probation end and the decision recorded when a probation ends (confirmed, extended to a new end date, or not confirmed, with a note and who recorded it), a direct manager's recommendation on a probation before it is decided (with an optional note for HR, which the employee is not shown), employment type, department (and who heads each department), job title, branch, shift, an own work week (the weekly days off) where you set one instead of the branch's, reporting manager, status, termination date, free-text notes. |
| Pay | Basic salary or hourly rate, allowances and deductions (with the months they are paid in), loans and advances, bank name and IBAN, payslips, the history of salary changes, and pay holds (the months held, the reason given and who set and released them). |
| Time | Attendance records, leave requests (with the approvers' and HR's notes on them), stated leave figures (the days carried over from your earlier records, extra days added to or taken off a year's allowance, and the days carried into and used in the year Peoplume began tracking a leave type, each with the reason given for the figure and who recorded it), timesheets and project hours (with the activity booked and the actions recorded on each timesheet: who, when and any comment). Portal and phone-app check-ins may carry the device's location where the employee's device allows it (the punch is never blocked if it does not), and then its distance from the branch's location as set at that moment, where the branch has one. |
| Expenses | Expense claims (title, category, amount, date, notes, receipt, and optionally a project; where you record tax, the tax inside the amount and the supplier's tax number; for a mileage claim, the distance entered and the rate applied), trip reports grouping claims (title, dates, purpose), and travel advances (amounts requested and approved, the reason given, the hand-over and the settlement). |
| Documents | Whatever you or your employees upload: employment contracts, identity documents such as passports and visas, certificates, expense receipts (uploaded in the portal or photographed in the phone app), candidate CVs, company policies. |
| Family | Dependents you record: name, relationship, optional date of birth and nationality, sponsorship, and one identity document's type, number and expiry date. |
| Performance and development | Goals (with notes, progress, the review cycle they belong to, and goals an employee proposes), review cycles and who each one covers (with the words a cycle gives each point of its rating scale), reviews and ratings (with the answers and scores given to each of the cycle's questions, and the employee's own comment on a review they have been shown), self-assessments (with a self-rating and answers to the cycle's questions), peer feedback, suggestions of colleagues to give peer feedback (with the reason and HR's decision), one-to-ones a manager records with a direct report (date, notes shared with the employee, the manager's private notes — never shown to the employee — and action items; deleted when the employee is erased), performance improvement plans (dates, objectives, linked goals, dated check-ins and the outcome; kept through erasure), praise (its notifications carry no text and are deleted with it), training enrolments, the sessions booked and the attendance recorded, certificate files, the reason an employee gives when asking for a course or for a certificate to be renewed (and the day HR declined a renewal request, when it did), succession plans (with the appointments made and the month a successor is expected to be ready). Engagement survey responses, stored with no name, no time and in an order unrelated to who answered, with the answers to a survey's extra statements stored the same way and unlinked to each other; the record of who responded keeps only the day. Where you break a survey's results down by branch or department, each answer is stored with the respondent's branch or department, never the person, but only for a group large enough to stand apart when the survey was created (at least the survey's minimum number of people who could answer it then); every other group's answers, a group created later's included, are pooled and stored with no group (when no group is too small, the smallest is pooled anyway, so the pool is never empty), and which groups are which does not change while the survey runs. |
| Recruitment | Job openings (title, department, branch, positions, the monthly salary range and its currency, why the seat is needed, the employee a hire replaces, and the careers page's closing date, city and whether the range is shown) and job requisitions (each request to open an opening or change its range, positions, branch or department: what it asked for, who sent and decided it, when, and the reason given for a decline). Hiring-team assignments; hiring managers' recommendations on candidates, with optional notes. Applicants, including those who apply through your public careers page: name, email, phone, CV, source (typed by you, or the channel the careers link was made for), the recruiter's notes, whether an applicant is an internal candidate (the employee record they already are), which employee referred them, and which earlier application they were put forward from; the candidate's request to be kept for future roles, its end and its withdrawal; and a dated history of the application (stage moves, why an application ended — withdrew, declined the offer, not a fit, position filled or other — with any later correction, notes added to it, corrections to the candidate's details before and after, interviews scheduled and their changes, offer changes, who made each, and when a rejection email was sent). |
| Interviews and offers | Interviews: date, time, the time zone they are read in, place, the interview panel (up to six colleagues, one the lead) and status; each panelist's own notes, overall rating and scorecard ratings, against the criteria fixed when the interview was scheduled. A self-scheduling link's note, the panel offered, who created it, the times offered and the time booked, and whether it was emailed. Offers: position, salary and its currency, start date, branch, notes, status, who sent and approved it, the letter wording approved and its respond-by date; offer-link facts (issued, expires, used, the answer) and a decline reason the candidate types — never their IP address or device. Accepting online is a record of the answer, not an electronic signature. |
| Candidate emails | Emails a recruiter writes to a candidate: subject, text, the address it went to, who sent it and when; the history records that one was sent, with its subject. Your own wording for candidate emails and the offer letter, and your recruiters' snippets. Emails to candidates name you and, when you set a recruiting contact address, carry it as their reply address; without one they ask the candidate to contact your hiring team instead of replying — except an email a recruiter writes, whose reply goes to the recruiter's own login email, which the recruiter is shown before sending. |
| Account and security | Login email, a bcrypt hash of the password, two-factor secrets where enrolled, last login time and the last sign-in through your identity provider; the sign-in history — when and how each sign-in was made, never a network address or device; counts of wrong passwords and of wrong two-step codes typed at sign-in, with when counting began and when the latest was typed (counting goes on until 30 days pass without a failure, and the next failure after that starts it again); a requested change of sign-in address (the new address, who asked and when) until it is confirmed, cancelled or replaced, or the login is removed; and an append-only audit log of sensitive changes and sign-ins recording who made them. For you as a company: the details you choose to print on payslips and letters (legal name, address, registration and tax numbers, and the name and title of the person who signs your letters), which an approved payroll run keeps as its payslips printed them, without the signatory. |
Special category data has no field in Peoplume, but two things can carry it anyway. Leave types are named by you, so a type called “Sick leave” makes the requests against it health data; and an uploaded document can be a medical certificate or anything else. Whether to record either is your decision, and Annex II describes the controls that exist around uploaded files. With the AI assistant switched on, leave rows reach Anthropic when a question looks them up: the leave-type names, and, for one person’s requests, the reason given and the approvers’ and HR’s notes on them. Engagement survey responses are the opposite case — they carry no employee reference and no timestamp, they are rewritten in a random order after every answer so their order in storage says nothing about who answered when, and the record of who responded keeps only the day. Results appear only once a survey is closed and at least its minimum number of people have answered, so no single answer can be watched arriving. A free-text comment can still identify its author; the portal says so before one is typed. A copy of the database holds no order or time that ties an answer to a person (each response keeps one storage key, fixed by its random identifier, through every rewrite, so anything the database engine leaves in unused space only repeats an answer under that same key, and that space is cleared every night; the space old rows used is overwritten when the responses are rewritten, and backups copy the current rows only), but two copies taken at different times — two nightly backups, say, or a disk image, which can also hold the temporary journal written while an answer is saved — can narrow it: comparing them shows the answers that arrived between them and who answered in between, so when only one person answered in that interval their answer is identified. That needs access to the database, its disk or its backups (see Annex II); no screen or export in the product offers it. While a survey is open, its count of people who could still answer is kept current, and someone coming off the list of people who can answer — a leaver, or a portal login removed — lowers it only if they had not answered, so a user watching the count across such a change can tell whether that person took part — not what they answered. A breakdown of a survey’s results by branch or department, which you choose when you create the survey, narrows anonymity further, and the survey form says so. An answer is stored with the respondent’s group only for a group large enough to stand apart when the survey was created — at least the survey’s minimum number of people who could answer it then; every other group’s answers (groups too small then, a larger group pooled beside them when they alone would be too few — or, when no group is too small, the smallest group anyway, so the pool is never empty — and any group created later) are pooled and stored with no group. Which groups are which is decided once, when the survey is created, and does not change while it runs. A group’s results appear only once at least the survey’s minimum number of people in it have answered, groups are held back until those not shown together hold at least the survey’s minimum (so the total cannot reveal them), a breakdown names only the groups it shows and then one line for all the others whatever was answered — a group not shown is never named, counted or given a figure — and comments are never broken down. In a small group, a shown figure can still say more about the few people in it than the whole survey’s does; and a copy of the database ties the answers stored under a group, the pooled ones included, to that group’s people who answered, so when only one of them did, to that person — a group large enough when the survey was created, the pool included, can still hold few answers, if its members have left or did not answer (again, access to the database, its disk or its backups).
Purpose. Running HR for you: employee records, attendance and shifts, leave, timesheets, expenses, payroll, performance, recruitment, training, the employee portal, notifications and exports.
Duration. Continuous while the account exists, then as section 7 says.
Competent supervisory authority. Per Clause 13 of the Standard Contractual Clauses: the authority of the Member State where you are established; or, where you are not established in the EEA but are subject to the GDPR under Article 3(2) and have appointed a representative, the authority where that representative is; or, failing that, the authority of a Member State where the data subjects are.
Annex II — Technical and organisational measures
This describes the product as it is today, including where it falls short. We hold no ISO 27001 or SOC 2 certification and have had no external penetration test. We would rather say so than let either be assumed. The fuller version of this, with the reasoning, is on the security page.
| Measure | What is actually implemented |
|---|---|
| Tenant separation | Every record carries the company it belongs to and every query filters on it; there is no shared pool a forgotten filter could expose. Company scope alone is not treated as sufficient — reads are also checked against the relationship that entitles you to them, such as your own record or your own report. Automated tests assert this specifically. |
| Encryption in transit | TLS on every connection, terminated by Cloudflare in front of the application. |
| Encryption at rest | Uploaded files and database backups are held in Cloudflare R2, which encrypts them at rest. We do NOT encrypt them ourselves before upload, so the storage provider holds them in a form it could read. The database file sits on the hosting provider's volume; we add no application-layer encryption to it. |
| Authentication | Passwords hashed with bcrypt and never stored or logged readably; a new password needs at least 8 characters and is refused when it is one of the 10,000 most common or contains the email address's name, the company's name or 'Peoplume'. Optional two-factor authentication (TOTP), which an administrator can require for every console user; a code works once. Single sign-on over OpenID Connect, set up and changed only by your Administrators, for email domains you have proven by a DNS record, and optionally required for console accounts on them (one named Administrator keeps a password); an enrolled account still gives its two-factor code after the provider. No SAML, deliberately. |
| Sessions | The session cookie is httpOnly, sameSite and secure, and carries only a user id — role, company, permissions and branch scope are re-derived from the database on every request, so a permission change takes effect on the holder's next request and nothing can be smuggled in the token. Changing or resetting a password, or 'sign out everywhere' (open to everyone, portal users included), invalidates every other session immediately, the phone app's included, and an administrator can sign a person out of all of theirs; each turns off push notifications on all of the account's devices. A browser sign-in lasts at most 7 days; you can shorten console sign-ins. A console account whose access has ended (an end date on the account, typically a leaver's) has no session at all; restoring it, or ending access on a day already past, signs it out of every browser and app device, so its holder signs in afresh. |
| Access control | Role-based permissions from a fixed server-side catalogue, with per-company roles and optional branch scoping that limits a user to one location's employees. Permissions are never taken from the request; a role can only grant what the person granting it holds. Candidate access in the employee portal is by assignment to an opening's hiring team, not by role, checked on every request, and ends when the person leaves the team, stops working here, or 90 days after the opening closes; nobody sees an opening they applied for. |
| Separation of duties | Nobody approves their own record anywhere in the product — leave, expenses, travel advances, timesheets, offers, compensation proposals, a job application of their own, a job requisition to replace themselves or an internal move of their own record. Expense claims above a threshold you set need a second, different approver; where you require it, opening a job or changing its salary range, positions, branch or department needs the approval of someone other than the person who asked, unless nobody else could approve it when it was sent. |
| Accountability | An append-only audit log records sensitive changes — salary and bank edits, job changes (department, job title, employment type, work week, manager, shift and branch), permission and role changes, approvals, deletions, data exports taken by staff about someone else or of a review cycle's results, candidate exports and deletions, the applicants spreadsheet, bulk stage moves (as counts), corrections to a candidate's details (names and source before and after; email, phone and CV only as changed), internal-candidate links, referrals, put-forwards to other openings and internal moves — with who, what and when, and every sign-in with how it was made (never a network address or device). Its pay figures show only to people who may view salaries, and a search of the log never matches them for anyone else. It can be filtered and downloaded by people allowed to view it; each download is recorded, and a download's pay figures follow the same rule. |
| Uploaded files | Private object storage, never a public URL. Every download goes through an authorised route that re-checks the session and the relationship: portal users can reach only their own documents, and a candidate's CV only while they sit on that candidate's interview panel or the opening's hiring team. The uploaded-files download needs permission to manage settings, and gives documents, receipts, CVs and training certificates only to someone with the permission of the page that shows them. Type and size are restricted on upload. |
| Data minimisation | Limited, and stated as such. Rejected job applicants can be purged on a schedule you set — with their interviews, feedback, offer, history, messages, the hiring manager's recommendation and the staff notifications that link to their application, except while a candidate's request to be kept for future roles lasts (never longer than your careers form promised, nor than your talent pool's current period, and not once you switch the pool off or withdraw the request); an interviewer's notice of an interview with them, notification emails already delivered and audit log entries naming them remain — and a former employee's personal data can be erased on request; otherwise data stays until someone deletes it. There is no general retention timer. |
| Resilience and recovery | Nightly backup with an integrity check and a checksum, copied off-site; the newest off-site copy is read back and checked each day. Restores are scripted and default to a dry run, so recovery does not begin by overwriting anything. |
| Abuse and availability | Rate limits on sign-in (web and phone app) per account, and per IP on failed attempts and more loosely on all attempts, with failures against existing accounts of companies not in a free trial or read-only also limited per email domain (webmail domains excepted) and globally, each account and, globally, each company counting only up to a share; on two-factor codes per account, per IP and, for wrong codes other than those typed while setting it up, globally in the same way; on sign-up, password reset, invitation acceptance and single sign-on per IP (reset emails capped globally); on adding users and sending invitations per person and per company, and on setting clock-in PINs and typing a new password for someone else per person; on the public application, kiosk, candidate-booking, job-feed and careers-logo endpoints and the offer response (per address before the link resolves; per link and overall after); and on emails written to candidates (at most 5 a day to one candidate and 200 a day per company — 20 during a trial — counted from the stored record). Password, PIN and new-password checks are also paced across the service, one pace per kind (sign-in, clock-in presses, other password steps), so a flood of one kind does not hold up the others, and each company's other password steps count only up to a share of their pace; at busy times a check waits a few seconds (at most three) or is asked to try again in a moment, and that answer is never counted as a wrong attempt. The IP is the one Cloudflare reports, an IPv6 address counted by its /64 block, and the server can be set to refuse requests that did not come through Cloudflare. API keys are stored only as a hash, carry explicit scopes limited to what their creator may see, and are rate limited per address and per key. Outgoing webhooks are HMAC-signed and refuse private, loopback and link-local addresses; incoming payment webhooks are signature-verified in constant time with a replay window. |
| Development practice | Every change runs a build, a type check and an automated test suite in continuous integration before release, including tests that assert tenancy and authorisation behaviour, and an end-to-end browser suite. |
| Browser hardening | Every response carries a Content-Security-Policy, HSTS for a year, nosniff, X-Frame-Options: DENY, a referrer policy and a permissions policy that disables camera, microphone, USB and motion sensors, disables payment on every page except the payment page (/billing/pay), where Paddle's checkout frame may use it, and permits location only for our own pages. Scripts are limited to our own origin and Cloudflare's analytics script, except that the payment page also loads Paddle.js from cdn.paddle.com and frames Paddle's checkout; inline script is still allowed, because the framework emits it, so the policy blocks loading an attacker's script but not executing an injected inline one. |
| Known gaps | No independent security certification or penetration test. Inline script is permitted, as above. No SCIM: turning someone off in your identity provider does not end their Peoplume account or a session they already have open. The public API is read-only. Survey answers stay anonymous against the product and — on a survey not broken down by branch or department — against a copy of the database, not against someone comparing two copies taken at different times or holding a disk image (Annex I); while a survey is open, its count shows whether someone who comes off the list of people who can answer (a leaver, or a login removed) had answered (Annex I). A survey broken down by branch or department shows each group's results once at least the survey's minimum number in it have answered, which narrows anonymity for a small group, and stores each answer with its group when the group was at least the survey's minimum size when the survey was created (pooling the smallest group anyway when no group was under that size, so the pool is never empty), so a copy of the database ties the answers stored under a group to that group's people who answered — to one person when only one of them did, which a group large enough then, the pool included, can still come to if its members leave or do not answer (Annex I). |
Annex III — Sub-processors
The current list, with what each one receives, is maintained on the sub-processors page, which is part of this annex. In summary:
| Sub-processor | Purpose | Location |
|---|---|---|
| Railway Corp. | Application hosting and the volume holding the database. | As deployed — ask before you commit if you have a residency requirement. |
| Cloudflare, Inc. | DNS and reverse proxy; R2 object storage holding every uploaded file and the off-site database backups; Web Analytics, a cookieless page-view counter that receives no employee data. | Global, distributed |
| GitHub, Inc. (Microsoft) | Runs the nightly database backup and the support commands a Peoplume owner runs outside the product, on GitHub-hosted runners. Receives each night's full database copy in transit and keeps it as a build artifact for 7 days, unencrypted by us; for a support command, its inputs and output (an account's email, a company's name and id, the names of the people it concerns, the reason recorded). Each day it also downloads the newest off-site copy to check that it opens, deleting it when the check ends. For a notice to Administrators: the notice's name, and in what the command prints the names and ids of companies and counts, never an Administrator's address. | United States |
| Paddle.com Market Ltd | Payments as merchant of record. Receives billing contact and payment details, and the number of seats billed (a count of employees who are not terminated, never a name); receives no employee data. | United Kingdom / EU |
| Resend, Inc. | Transactional email — the link that confirms a sign-up, notifications, invitations, password resets, links confirming a new sign-in address, security notices, service notices to Administrators, messages to candidates (including a self-scheduling link when a recruiter emails it, and an offer's response link). Used for every company: sign-up cannot be completed without it. Each person can turn off email copies of their notifications; security notices ignore that setting. So do Peoplume's service notices to a company's Administrators (a new sub-processor, a breach). Notification copies, the monthly digest and candidate emails carry your company's name, followed by “via Peoplume”, as the sender's name; account and security emails are sent as Peoplume. | United States |
| Google LLC, Apple Inc., Mozilla Corporation (push services) | Relay push notifications to devices whose user turned them on. The message is encrypted to the device before it leaves us; the service learns only that a message was sent to that subscription. | United States / global |
| 650 Industries, Inc. (Expo) | Delivers notifications to the Peoplume phone app via Apple's and Google's push services, for phones where the employee signed in and allowed notifications. Receives the phone's push token and a generic alert (for example “Leave update”) with an in-app link; never a name, date, amount or free text, because this message is not encrypted to the device. The link names the page only where that says nothing sensitive: for a one-to-one, an improvement plan or a probation it is the app's own inbox, and the app fetches the real page from Peoplume after the tap. | United States |
| Anthropic PBC | Receipt reading and the AI assistant, each off until an administrator switches it on. Receives one receipt image, with the company's expense category names, when someone presses the button for it, only while an administrator has receipt reading switched on; or, only while an administrator has the assistant switched on, a question with the asker's name, the company's name and today's date, the earlier questions and answers of the same conversation (up to 20 messages, as text), and the HR rows looked up to answer it. Anthropic handles it under its own Commercial Terms of Service and Data Processing Addendum, not terms Peoplume negotiated. Peoplume does not promise that Anthropic will not use it to train or improve its models, or keep what it uses for that. Anthropic's published policy for API customers is to delete what it receives within 30 days, except content its automated safety systems flag under its Usage Policy (kept for up to 2 years, with the safety classification scores for up to 7 years) and anything the law requires it to keep longer. Peoplume has no zero-data-retention arrangement with Anthropic. To reduce cost, each assistant request is cached for about five minutes after its last use (a receipt request is never cached); Anthropic says the cache holds an internal representation of the prompt in memory, not the question or the answer themselves. Peoplume stores no conversation. | United States |
Administrators on the account are notified before this list changes, as section 5 says. Job boards and search engines that fetch a job feed or careers page you publish receive only the job details you chose to publish, at your instruction; they are not our sub-processors.
Contact
Questions about this agreement, or to ask for a signed copy: [email protected]. Data protection enquiries: [email protected].
