@ever-co/legal 0.5.0 → 0.6.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (103) hide show
  1. package/dist/content/aup/ever-async/en.json +2 -2
  2. package/dist/content/aup/ever-corporate/en.json +2 -2
  3. package/dist/content/aup/ever-demand/en.json +2 -2
  4. package/dist/content/aup/ever-rec/en.json +2 -2
  5. package/dist/content/aup/ever-teams/en.json +2 -2
  6. package/dist/content/aup/ever-works/en.json +2 -2
  7. package/dist/content/aup/everiq/en.json +2 -2
  8. package/dist/content/aup/gauzy/en.json +2 -2
  9. package/dist/content/aup/githands/en.json +2 -2
  10. package/dist/content/aup/hust/en.json +2 -2
  11. package/dist/content/aup/traduora/en.json +2 -2
  12. package/dist/content/cookies/ever-async/en.json +2 -2
  13. package/dist/content/cookies/ever-corporate/en.json +2 -2
  14. package/dist/content/cookies/ever-demand/en.json +2 -2
  15. package/dist/content/cookies/ever-rec/en.json +2 -2
  16. package/dist/content/cookies/ever-teams/en.json +2 -2
  17. package/dist/content/cookies/ever-tech/en.html +11 -3
  18. package/dist/content/cookies/ever-tech/en.json +4 -4
  19. package/dist/content/cookies/ever-works/en.json +2 -2
  20. package/dist/content/cookies/everiq/en.json +2 -2
  21. package/dist/content/cookies/gauzy/en.json +2 -2
  22. package/dist/content/cookies/githands/en.json +2 -2
  23. package/dist/content/cookies/hust/en.json +2 -2
  24. package/dist/content/cookies/traduora/en.json +2 -2
  25. package/dist/content/dpa/ever-async/en.json +2 -2
  26. package/dist/content/dpa/ever-demand/en.json +2 -2
  27. package/dist/content/dpa/ever-rec/en.json +2 -2
  28. package/dist/content/dpa/ever-teams/en.json +2 -2
  29. package/dist/content/dpa/ever-tech-services/en.html +1847 -0
  30. package/dist/content/dpa/ever-tech-services/en.json +624 -0
  31. package/dist/content/dpa/ever-works/en.json +2 -2
  32. package/dist/content/dpa/gauzy/en.html +5 -10
  33. package/dist/content/dpa/gauzy/en.json +3 -3
  34. package/dist/content/dpa/githands/en.json +2 -2
  35. package/dist/content/dpa/hust/en.json +2 -2
  36. package/dist/content/dpia/gauzy/en.json +2 -2
  37. package/dist/content/privacy/ever-async/en.json +2 -2
  38. package/dist/content/privacy/ever-corporate/en.json +2 -2
  39. package/dist/content/privacy/ever-demand/en.json +2 -2
  40. package/dist/content/privacy/ever-rec/en.json +2 -2
  41. package/dist/content/privacy/ever-teams/en.json +2 -2
  42. package/dist/content/privacy/ever-tech/en.html +55 -12
  43. package/dist/content/privacy/ever-tech/en.json +12 -4
  44. package/dist/content/privacy/ever-works/en.json +2 -2
  45. package/dist/content/privacy/everiq/en.json +2 -2
  46. package/dist/content/privacy/gauzy/en.json +2 -2
  47. package/dist/content/privacy/githands/en.json +2 -2
  48. package/dist/content/privacy/hust/en.json +2 -2
  49. package/dist/content/privacy/traduora/en.json +2 -2
  50. package/dist/content/refund/ever-corporate/en.json +2 -2
  51. package/dist/content/refund/ever-demand/en.json +2 -2
  52. package/dist/content/refund/ever-rec/en.json +2 -2
  53. package/dist/content/refund/ever-teams/en.json +2 -2
  54. package/dist/content/refund/ever-tech-services/en.html +270 -0
  55. package/dist/content/refund/ever-tech-services/en.json +178 -0
  56. package/dist/content/refund/ever-works/en.json +2 -2
  57. package/dist/content/refund/gauzy/en.json +2 -2
  58. package/dist/content/refund/githands/en.json +2 -2
  59. package/dist/content/refund/hust/en.json +2 -2
  60. package/dist/content/refund/traduora/en.json +2 -2
  61. package/dist/content/security/ever-async/en.json +2 -2
  62. package/dist/content/security/ever-corporate/en.json +2 -2
  63. package/dist/content/security/ever-demand/en.json +2 -2
  64. package/dist/content/security/ever-rec/en.json +2 -2
  65. package/dist/content/security/ever-teams/en.json +2 -2
  66. package/dist/content/security/ever-tech/en.json +2 -2
  67. package/dist/content/security/ever-works/en.json +2 -2
  68. package/dist/content/security/gauzy/en.json +2 -2
  69. package/dist/content/security/githands/en.json +2 -2
  70. package/dist/content/security/hust/en.json +2 -2
  71. package/dist/content/security/traduora/en.json +2 -2
  72. package/dist/content/subprocessors/ever-async/en.json +2 -2
  73. package/dist/content/subprocessors/ever-corporate/en.json +2 -2
  74. package/dist/content/subprocessors/ever-demand/en.json +2 -2
  75. package/dist/content/subprocessors/ever-rec/en.json +2 -2
  76. package/dist/content/subprocessors/ever-teams/en.json +2 -2
  77. package/dist/content/subprocessors/ever-tech/en.html +16 -2
  78. package/dist/content/subprocessors/ever-tech/en.json +4 -4
  79. package/dist/content/subprocessors/ever-tech-services/en.html +357 -0
  80. package/dist/content/subprocessors/ever-tech-services/en.json +158 -0
  81. package/dist/content/subprocessors/ever-works/en.json +2 -2
  82. package/dist/content/subprocessors/everiq/en.json +2 -2
  83. package/dist/content/subprocessors/gauzy/en.json +2 -2
  84. package/dist/content/subprocessors/githands/en.json +2 -2
  85. package/dist/content/subprocessors/hust/en.json +2 -2
  86. package/dist/content/subprocessors/traduora/en.json +2 -2
  87. package/dist/content/tos/ever-async/en.json +2 -2
  88. package/dist/content/tos/ever-corporate/en.json +2 -2
  89. package/dist/content/tos/ever-demand/en.json +2 -2
  90. package/dist/content/tos/ever-rec/en.json +2 -2
  91. package/dist/content/tos/ever-teams/en.json +2 -2
  92. package/dist/content/tos/ever-tech/en.html +46 -11
  93. package/dist/content/tos/ever-tech/en.json +12 -4
  94. package/dist/content/tos/ever-tech-services/en.html +937 -0
  95. package/dist/content/tos/ever-tech-services/en.json +473 -0
  96. package/dist/content/tos/ever-works/en.json +2 -2
  97. package/dist/content/tos/everiq/en.json +2 -2
  98. package/dist/content/tos/gauzy/en.json +2 -2
  99. package/dist/content/tos/githands/en.json +2 -2
  100. package/dist/content/tos/hust/en.json +2 -2
  101. package/dist/content/tos/traduora/en.json +2 -2
  102. package/dist/index.json +1788 -343
  103. package/package.json +52 -52
@@ -8,7 +8,7 @@
8
8
  "locale": "en",
9
9
  "version": "1.0.3",
10
10
  "effectiveDate": "2026-10-24",
11
- "sha256": "d6008ff6bc5823dc8af4614129a08e66f971d92bd78673288da4e1d26dd356e3",
11
+ "sha256": "91dc6e6a9e3d02673e818485a00d04b193e71c147d691c864fad49bbb6d160aa",
12
12
  "publishable": true,
13
13
  "sections": [
14
14
  "core/security/010-overview",
@@ -376,5 +376,5 @@
376
376
  }
377
377
  ],
378
378
  "path": "content/security/gauzy/en.html",
379
- "markdown": "## Security overview\n\nThis page describes how we protect Ever Gauzy Cloud and the data you keep in it. It is written to be\nchecked rather than admired: every measure below is something we actually run, and where we do not do\nsomething that a page like this usually claims, **we say so instead of leaving a gap you have to guess at**.\n\nThe short version:\n\n- We run the Service on **hardware we own and operate ourselves**, not as a tenant in someone else's cloud.\n The database, the object storage and the secrets store are ours. That is why the list of providers with\n access to your data is short.\n- Traffic is encrypted in transit, storage is encrypted at rest, and off-site backup copies are encrypted\n **before they leave our network**, so the provider holding them holds ciphertext.\n- Production changes go through version control and review. Hand edits on live systems are the exception,\n not the routine.\n- **We hold no security certification, and we do not claim one.** The next section says exactly what we do\n instead.\n\n### What this page covers\n\nIt covers the hosted Service we operate at [gauzy.co](https://gauzy.co) — the application, its\ninterfaces, and the infrastructure underneath them.\n\nIt does **not** cover two things that are easy to confuse with it:\n\n- **A copy of the software you run yourself.** Where our software is published as open source and you deploy\n it on your own infrastructure, you are the operator. You choose its hosting, its storage and every other\n component, we have no access to it, and nothing on this page describes it.\n- **Your own systems and your own configuration.** Which of your people have access, how you configure the\n product, what you connect it to and how you secure the devices your team uses are yours. The shared\n responsibility section near the end of this page sets out that split in detail, because most real\n incidents happen on one side of that line rather than in the middle of it.\n\n### What this page is, legally\n\nThis is a factual description of the measures in place at the version and date shown at the end. It is\n**not a warranty, and it does not create a service level commitment.** Our contractual security obligations\nare in our Data Processing Addendum and in the [Terms of Service](https://gauzy.co/tos), and those are the\ndocuments that bind us.\n\nWe change these measures as the systems and the threats change. **We will not make a change that materially\nreduces the protection described here**, and when the description changes we republish this page with a new\nversion and date.\n\n### One thing we will not tell you\n\n**No system is perfectly secure, and a vendor who tells you otherwise is selling something.** Everything\nbelow reduces risk; none of it eliminates it. What we can commit to is that we take these measures\nseriously, that we test the ones that can be tested, and that when something goes wrong we tell you rather\nthan manage the news.\n\n## Certifications — what we hold, and what we do not\n\n**We do not hold a SOC 2 report of any type. We are not certified to ISO/IEC 27001. We hold no other\nsecurity certification or attestation, and we do not claim one.**\n\nWe are stating that first, in plain words, because a security page that leaves the question open is doing\nso on purpose. If a proposal, a sales conversation, a questionnaire response, a directory listing or a\nthird-party summary has told you that we are SOC 2 or ISO 27001 certified, **it is wrong** — please send it\nto legal@ever.co so we can get it corrected at the source.\n\n### What we do instead\n\nWe use two published control sets as a working checklist rather than as a badge:\n\n- **The control areas of ISO/IEC 27001 Annex A** — access control, cryptography, operations security,\n supplier relationships, incident management, continuity — as the structure for reviewing our own\n practices. This page is organised broadly along those lines.\n- **Article 32 GDPR**, which is not optional for us and is the standard our customers can actually hold us\n to: technical and organisational measures appropriate to the risk, kept under review as the risk changes.\n\n**No external auditor has examined either.** These are self-assessments, and you should read them as such.\nAn unaudited framework alignment and a certification are different things, and we will not blur the two by\nwriting \"aligned with\" in a way that reads like \"certified to\".\n\n### What you can have instead of a certificate\n\nA certificate is a proxy for evidence. Where we cannot give you the proxy, we will give you the evidence:\n\n- **This page**, which is specific enough to be checked against what we actually do.\n- **The security measures schedule in our Data Processing Addendum**, which is contractual rather than\n descriptive — if a measure matters to you, that is where to pin it down.\n- **A completed security questionnaire.** Send yours to legal@ever.co. We answer them, we answer them\n honestly, and where the honest answer is \"no\" you will get \"no\" rather than a paragraph that avoids the\n question.\n- **Evidence on request** for specific points — a description of the architecture, the outcome of our most\n recent restore test, the current sub-processor list, the transfer mechanism for a named provider.\n- **A right to audit**, on the terms set out in the Data Processing Addendum.\n\n### If that changes\n\nIf we obtain a certification or an independent report, this page will say which one, who issued it, what\nits scope covers and what it excludes — because the scope is the part that matters and the part most often\nleft out. Until that day, the sentence at the top of this section is the whole answer.\n\n## Encryption\n\n### In transit\n\nAll traffic between you and the Service travels over TLS. HTTPS is enforced and plain HTTP requests are\nredirected rather than served, so a request that arrives unencrypted never reaches the application with its\ncontents intact. We keep the accepted protocol versions and cipher suites current and retire weak ones as\nthe ecosystem does.\n\nTraffic between our own sites, and between our infrastructure and the edge network in front of it, runs\nover encrypted connections. Traffic to the providers we engage runs over TLS to their published endpoints.\n\n### At rest\n\n- **Storage holding customer content, databases and backups is encrypted at rest.** That protects against\n the loss, theft or improper disposal of physical media — it does not protect against a compromise of a\n running system that already has the data decrypted, and we would rather you knew which threat it answers.\n- **Off-site backup copies are encrypted before they leave our network**, with keys we hold and manage\n ourselves. The provider storing those copies holds ciphertext and nothing else: it cannot read them, and\n neither can anyone who obtains them from it.\n\n### At the application layer\n\nSome values are too sensitive to rely on storage-level encryption alone, so they are encrypted again inside\nthe application before they are written:\n\n- **Credentials, access tokens and API keys** — including tokens you grant us for a third-party integration.\n- **Secrets used to operate the Service** — held in a secrets manager we run ourselves, injected into\n workloads at runtime, and **never committed to source control**.\n\n**Passwords are never stored in a form we can read.** They are put through a modern one-way hashing\nfunction with a per-password salt. We cannot recover your password, tell you what it is, or send it back to\nyou — a reset is the only route, and any service that offers you the alternative is storing it wrongly.\n\nWhere a specific product stores a particular credential differently from the description above, its annex\nto this page says so and says why. **We would rather publish an awkward exception than a tidy sentence that\nis not true everywhere.**\n\n### Keys\n\nEncryption keys are held in our self-hosted secrets manager, separately from the data they protect, and are\naccessible only to the systems and people that need them. Keys are rotated when a person with access leaves,\nwhen we have reason to think a key has been exposed, and on a periodic schedule. **The keys for off-site\nbackups are held independently of the infrastructure they protect**, so that losing a site does not also\nlose the ability to decrypt its backups.\n\n## Access control\n\n### Your access to your own data\n\nAccess inside your workspace is yours to configure. The Service gives you named user accounts, roles and\npermissions, workspace administrators who can invite and remove people, and session controls. Where a\nproduct supports single sign-on, multi-factor authentication or IP restrictions, its annex says so and its\ndocumentation says how to switch them on.\n\n**Turning those controls on, and reviewing who still needs access, is your job — see the shared\nresponsibility section.** The most common way an account is compromised is not a flaw in our systems; it is\na shared password, a former employee who was never removed, or an integration token that outlived its\npurpose.\n\n### Our access to production systems\n\n- **Named individual accounts, never shared logins.** Every action in production is attributable to a\n person.\n- **Multi-factor authentication** on the systems and administrative accounts that support it.\n- **Least privilege.** Access is scoped to the narrowest role that lets the work happen, and elevated\n access is granted for a purpose rather than held permanently by default.\n- **Changes go through Git.** The configuration of the live environment is held in version control and\n applied automatically, so a change is reviewable, attributable and reversible. Direct hands-on changes to\n a running system are an exception for incidents and hands-on maintenance, not the daily routine — and the\n automation reverts undeclared edits rather than leaving them in place.\n- **Removal on departure or role change.** Access is revoked when someone leaves or no longer needs it, and\n credentials they held are rotated. Access is reviewed periodically as well, because revocation lists rot.\n\n### Access to your content\n\nWe access the content in your workspace only where it is necessary to operate, secure or support the\nService — for example to investigate a fault you have reported, or to respond to a security incident. **We\ndo not browse customer content, and we do not use it to build products, train models or produce insights\nfor anyone else.**\n\nSupport access is granted for the task in hand and logged. Where a product offers a mechanism for you to\ngrant or revoke support access yourself, its annex describes it.\n\n### Being honest about the limits\n\n**We are a small team, and several of us hold administrative access because operating this infrastructure\nrequires it.** A large organisation would separate the person who approves a change from the person who\nmakes it; at our size that separation is partial, and we are not going to describe it as complete. What we\ndo instead is make every change visible: production state comes from version control, actions are\nattributable to named accounts, and administrative events are logged and monitored.\n\n## Infrastructure and hosting\n\n### We run our own infrastructure\n\nThe Service does not run in a rented public cloud account. It runs on **physical servers we own, in\nfacilities we control, inside the European Union**, on a virtualisation and container platform we operate\nourselves. The database, the object storage, the message and cache layers, the secrets store and the\ndeployment system are all components we run — not managed services bought from a third party.\n\nThat is an unusual answer in this market, and it has a direct consequence you can verify: **the list of\nproviders with access to your data is short.** Most of what would otherwise be a sub-processor is simply a\nservice we operate.\n\nIn outline, and deliberately without the detail that would help someone attack it:\n\n- **Hardware and virtualisation we own**, clustered so that losing a machine does not lose a service.\n- **A container orchestration platform** running the applications, with the control plane replicated across\n multiple machines.\n- **PostgreSQL**, run by an operator that handles replication, failover and continuous archiving of the\n write-ahead log.\n- **S3-compatible object storage** we host, for files and media, with data replicated across nodes.\n- **A secrets manager we host**, from which credentials are injected into workloads at runtime.\n- **Declarative configuration held in Git** and applied automatically, so the live environment matches a\n reviewed commit rather than an accumulation of manual changes.\n\nA content delivery and security network sits in front of all of it, and our source code, build pipeline and\ncontainer registry are hosted with a third-party provider. Those are genuine third parties even though the\ninfrastructure behind them is ours, and they are named on our\n[sub-processor page](https://gauzy.co/subprocessors).\n\n### Redundancy and failure domains\n\nServices that can run on more than one machine run on more than one machine. Databases are replicated,\nobject storage is replicated across nodes, and the systems that hold backups run on **a physically separate\ncluster from the systems they protect** — so a failure that takes out the live environment does not take\nits backups with it.\n\n**This is not the same as multi-region redundancy, and we will not describe it as such.** We can lose\nmachines without losing data or availability. A whole-site failure means recovering from off-site copies,\nwhich is a matter of hours rather than seconds. The continuity section is specific about what that means.\n\n### Separation between customers\n\nIsolation between customers is **logical, not physical**, unless you have separately contracted for a\ndedicated deployment. Depending on the product that means a separate database per customer, a separate\nnamespace and deployment per customer site, or tenant scoping enforced in the application and the data\nmodel. **Your product's annex states which model applies to you** — it is a question worth asking of any\nvendor, and worth a straight answer.\n\n### Separation between environments\n\nDevelopment, staging and production are separate environments with separate credentials and separate data\nstores. Production credentials are not available in development or staging, and **we do not copy production\ndata into a development environment**. Test data is synthetic or anonymised.\n\n### Third-party managed services, where they are used\n\nA small number of products use a third-party managed database, managed storage or hosted platform for a\nspecific function rather than our own. Where that is the case it is named in that product's annex and it\nappears in the always-engaged tier of the sub-processor list. **We do not describe the estate as fully\nself-hosted where a product's data actually sits somewhere else.**\n\n## Network security\n\n### At the edge\n\nEvery request to the Service passes through a content delivery and security network before it reaches us.\nThat layer terminates TLS, absorbs volumetric denial-of-service traffic, applies a web application firewall\nand filters automated abuse. It also means the address of our origin infrastructure is not the address your\nbrowser connects to.\n\n### The origin is not exposed\n\n**Our application servers do not publish inbound ports to the public internet.** Traffic reaches them\nthrough outbound-initiated tunnels established from inside our network to the edge, so there is no\npublic listener to scan, and no direct path from the internet to an origin host.\n\n### Inside the network\n\n- Databases, object storage, the secrets manager and internal services sit on **segmented private networks\n and are not reachable from the internet**. They accept connections only from the workloads that need them.\n- **Administrative interfaces are not published publicly.** Reaching them requires being on the internal\n network, through a controlled route with authenticated access.\n- Outbound connectivity from workloads is limited to what the workload actually needs, so a compromised\n component has fewer places to send data and fewer places to fetch a second stage from.\n\n### Abuse and automation\n\nSign-up, sign-in, password reset and contact endpoints are protected against automated abuse by rate\nlimiting and by a bot-detection challenge. **That challenge is strictly necessary to keep the Service\nworking and available** — it protects accounts from credential-stuffing and forms from bulk abuse — which\nis why it runs without a consent prompt, as explained in our [Cookie Policy](https://gauzy.co/cookies). It\nis not used for analytics, profiling or advertising.\n\n### What we monitor for\n\nUnusual authentication patterns, unexpected administrative activity, sudden changes in traffic shape and\nfailures of the components above raise alerts that a person sees. The logging and monitoring section\ndescribes what is recorded and for how long.\n\n## Application security\n\n### How a change reaches production\n\nEvery change to the Service starts as a change to source code in version control and follows the same\nroute:\n\n1. **A pull request**, reviewed by someone other than its author on the branches that matter.\n2. **Automated checks** — build, tests, static analysis and dependency scanning — which must pass.\n3. **Promotion through environments.** Changes move development to staging to production, in that order.\n Production is deployed from the release branch and from nothing else.\n4. **A container image built by the pipeline** from a reviewed commit, identified by its digest, deployed\n by the same automation that holds the environment's configuration in Git.\n\n**Nobody edits production by hand as a way of shipping a change.** The deployment system reconciles the\nlive environment against the version-controlled definition, so an undeclared edit is reverted rather than\nquietly kept.\n\n### Automated checks on every change\n\n- **Dependency scanning**, with alerts for known vulnerabilities in the packages we use, and updates\n applied on a defined schedule with a faster path for serious issues.\n- **Static analysis** of our source code, run on each change, flagging the classes of defect it can find\n before the code ships.\n- **Secret scanning** across our repositories, with pushes blocked where a credential is detected. Where a\n credential has been exposed, we treat rotating it at the provider as the fix — deleting it from the code\n is not the fix.\n- **Automated tests**, which have to pass before a change can be merged or released.\n\nDependencies are installed through an internal package cache we run, which gives us a consistent, auditable\nview of what actually went into a build.\n\n### In the application itself\n\n- Database access goes through parameterised queries and a mapping layer rather than assembled strings.\n- Output is encoded for its context, and state-changing requests are protected against cross-site request\n forgery.\n- Authentication, session handling and password storage use established libraries rather than anything we\n invented; passwords are hashed as described in the encryption section.\n- Authorisation is enforced on the server for every request. A hidden control in the interface is not an\n access control.\n- Input is validated at the boundary, and uploads are constrained by type and size.\n- Browser-facing responses carry security headers, including a content security policy on the surfaces\n where one can be applied without breaking the product.\n\n### The honest limit\n\n**Automated tooling finds categories of defect, not all defects.** Review catches more, and neither catches\neverything. That is why the sections on logging, incident response and responsible disclosure exist: they\nare what we rely on for the bugs that get through, and something always gets through.\n\n## Testing\n\n### What we test, and how often\n\n- **Automated test suites** run on every change, in the pipeline, before anything can be merged or\n released.\n- **Static analysis and dependency scanning** run on every change as well, as described in the previous\n section.\n- **Restore tests.** We restore backups into a scratch environment and check the result against the source\n rather than assuming a backup that completed is a backup that works. This is the test most often skipped\n in the industry and the one most often needed.\n- **Manual review** of changes that touch authentication, authorisation, payment flows, data export or the\n handling of credentials — the areas where an automated check is least likely to notice that the logic is\n wrong.\n- **Configuration review** of the version-controlled definition of the live environment, which is where a\n surprising share of real exposure lives.\n\n### Penetration testing — what we do not have\n\n**We have not commissioned an independent penetration test of the hosted Service, and we do not claim one.\nThere is no test report for us to send you, and any summary suggesting otherwise is not ours.**\n\nWe would rather write that sentence than the usual formulation about \"regular security assessments\" that\nturns out on inspection to mean an automated scanner. When we do commission a test, this page will say who\nperformed it, when, what was in scope and what was excluded.\n\n### Testing the Service yourself\n\n**You may not run security testing against the Service without our written agreement** — automated\nscanning, load testing, fuzzing and exploitation attempts included. From our side an unannounced scan is\nindistinguishable from an attack, and it will be handled as one.\n\nIf you want to test, we will usually say yes. Write to legal@ever.co and we will agree the scope, the\nwindow, the accounts you use and how findings are reported. The conditions are the obvious ones: your own\ntest data only, no other customer's data touched, nothing that degrades the Service for anyone else, and\nfindings to us before they go anywhere else.\n\nIf you have already found something without meaning to, do not test further — go straight to the\nresponsible disclosure section, which is written for exactly that situation and offers you protection for\nit.\n\n## Logging and monitoring\n\n### What we record\n\n- **Authentication events** — sign-in, sign-out, failed attempts, password and multi-factor changes.\n- **Administrative and security-relevant events** — permission changes, invitations and removals, token\n creation and revocation, changes to workspace settings, access to production systems.\n- **Application programming interface access** — the request, its source, its outcome and its timing.\n- **Infrastructure and platform events** — deployments, configuration changes, node and service state.\n- **Application errors and diagnostics** — enough to work out what broke.\n\n### What we try not to record\n\nWe do not deliberately log passwords, session tokens, API keys or payment card numbers, and where such a\nvalue would otherwise appear in a log or an error report it is redacted. **We will not claim this is\nperfect** — a value can end up in a stack trace nobody predicted. When we find one, we treat it as an\nexposure: the credential is rotated and the log entry removed.\n\nWhere a product sends error reports to a third-party error-tracking service, that service is named in the\nproduct's annex and on the [sub-processor page](https://gauzy.co/subprocessors), because an error report can\ncarry an identifier for the person who hit the error.\n\n### How long we keep it\n\n| Log type | Retention |\n|---|---|\n| Security, access and audit logs | **12 months** |\n| Application and error logs, crash reports, diagnostics | **90 days** |\n\nThese are the same periods published in the retention table of our\n[Privacy Policy](https://gauzy.co/privacy), and if the two ever disagree the Privacy Policy governs.\n\nAccess to logs is restricted to the people whose job needs it, and log access is itself an administrative\nevent that gets recorded.\n\n### Monitoring and alerting\n\nMetrics from the platform, the databases, the storage layer and the applications feed a monitoring system\nthat raises alerts when something is wrong — a service failing, a certificate close to expiry, a backup\nthat did not complete, a disk filling, an unusual pattern of authentication failures. **Alerts go to people,\nnot to a dashboard nobody opens.**\n\nWe also run an **independent external check** that watches the Service and the monitoring system itself\nfrom outside our network, and raises an alarm if either goes quiet. Monitoring that cannot survive the\nfailure it is meant to detect is decoration, and a failed backup that alerts nobody is the standard way a\nrecovery turns out to be impossible.\n\n## Incident response\n\n### How we handle an incident\n\n1. **Detect and record.** An alert, a customer report or a disclosure from a researcher opens an incident,\n with one named person responsible for running it.\n2. **Assess and classify.** What happened, what is affected, whether personal data is involved, whose data,\n and how bad it could get. Severity drives everything that follows.\n3. **Contain.** Cut the access, revoke the credential, isolate the component, block the route — stopping it\n comes before understanding it.\n4. **Eradicate and recover.** Remove the cause, restore service, restore data from backup where that is\n what is needed, and verify the result rather than assuming it.\n5. **Notify.** Under the timescales below.\n6. **Review afterwards.** A written review of what happened, why, what we changed, and what would have\n caught it sooner.\n\n### Telling you about it\n\n**Where we are the controller** — for the data described in our [Privacy Policy](https://gauzy.co/privacy) as\nours — and there is a personal data breach that is likely to result in a risk to people:\n\n- We notify the CPDP **within 72 hours of becoming aware of it**, as\n Article 33 GDPR requires, and where we cannot give complete information in that window we give what we\n have and follow up.\n- We notify the affected people **without undue delay** where the breach is likely to result in a high risk\n to them.\n\n**Where we are your processor** — everything inside your workspace, including data your organisation\ncollects about its own personnel — you are the controller and the notification duty to the authority is\nyours, so ours is to you:\n\n- We notify you **without undue delay after we become aware, and in any event within 72 hours.**\n- We give you what Article 33(3) requires you to have: the nature of the breach, the categories and\n approximate number of people and records involved, the likely consequences, the measures we have taken or\n propose, and a contact point who can answer follow-up questions.\n- We assist you with your own notifications and with any questions from your supervisory authority. Our\n Data Processing Addendum sets that out contractually.\n\nWe notify you at the administrative contact addresses on your account. **Keeping those current is worth\nfive minutes of your time**, because a notification that reaches a mailbox nobody reads has been made and\nnot received.\n\n### What we will not do\n\n- **We will not sit on a breach** to work out how it looks, or wait until we have a complete picture before\n telling you something is wrong.\n- **We will not require you to sign anything** — a non-disclosure agreement, a waiver, an amended contract —\n as a condition of being told.\n- **We will not describe an incident as something else.** If data was accessed, the notice says data was\n accessed.\n\n### Reporting an incident to us\n\nIf you think an account has been compromised, credentials have leaked, or you have seen something in the\nService you should not be able to see, tell us immediately at abuse@ever.co. If personal data is\ninvolved, copy privacy@ever.co. Both addresses are monitored and reach the people who operate the\nService.\n\nIf you have found a vulnerability rather than an incident in progress, the responsible disclosure section\nbelow is the route, and it offers you protection that this section does not.\n\n## Continuity and backups\n\n### Three tiers, and why there are three\n\nBackups fail in three different ways — the copy is too old, the copy is on the machine that died, or the\ncopy is in the building that burned. So there are three tiers, each answering a different failure:\n\n- **Tier 1 — a fast local copy.** Recent database backups are staged on storage separate from the live\n database, on a different machine. This is the tier that answers \"somebody dropped the wrong table twenty\n minutes ago\" and it restores quickly because the data never left the site.\n- **Tier 2 — a durable on-site copy.** Everything is copied into a **private, backup-only object store that\n runs on a physically separate cluster from the systems it protects**, with its own credentials, reachable\n only from inside our network and never from the internet. Losing the cluster that runs the Service does\n not lose the cluster that holds its backups.\n- **Tier 3 — an off-site copy with a third party.** A further copy goes to object storage operated by an\n external provider, in a different place entirely, **encrypted before it leaves our network**. The\n provider stores ciphertext and holds no key. This is the tier that answers the loss of the whole site.\n\nDeliberately, **the credentials that run the Service are not the credentials that can delete its backups.**\n\n### What is backed up\n\n- **Databases** — continuous archiving of the write-ahead log for point-in-time recovery, plus periodic\n full backups and separate logical dumps. Two mechanisms, because they fail differently.\n- **Object storage** — the files and media held for you, mirrored to the backup tiers.\n- **Cluster and platform state**, including the configuration needed to rebuild the control plane.\n- **The secrets store**, snapshotted on a short cycle.\n- **The version-controlled configuration** that defines the environment, which is what makes a rebuild\n possible rather than merely theoretical.\n\n**Not backed up, on purpose:** caches and queues, which are disposable by design and reconstructed on\nrestart, and build artefacts that the pipeline can regenerate. Backing those up would add cost and\nrestore time for nothing.\n\n### Recovery objectives\n\nThese are the objectives we design and test against. **They are internal targets, not contractual\ncommitments** — no service level applies unless we have separately agreed one with you in writing, as the\n[Terms of Service](https://gauzy.co/tos) says.\n\n| Scenario | Recovery point — how much data is at risk | Recovery time — how long it takes |\n|---|---|---|\n| Loss of a machine or a service | None. Replicated components carry on | Minutes, usually without anyone noticing |\n| Database corruption or an unwanted change | Minutes — continuous archiving supports point-in-time recovery | Hours, depending on the size of the database |\n| Loss of the storage holding files | Last completed replication cycle | Hours |\n| **Loss of the whole site** | Last completed off-site cycle | **Hours to days.** This is the honest number |\n\n**We are candid about the last row.** We are resilient to losing machines; we are not a multi-region\nservice that fails over silently, and we will not describe ourselves as one.\n\n### We test restores\n\nA backup nobody has restored is a guess. We restore into a scratch environment and verify the result\nagainst the source — record counts, integrity checks, and whether the restored system actually starts —\nrather than trusting a job that reported success. Restore drills are documented, and the outcome of the\nmost recent one is available to customers on request at legal@ever.co.\n\nMonitoring alerts on a backup that is late, failing or stale, because **a backup system that fails silently\nis functionally the same as no backup system**, and is discovered at the worst possible moment.\n\n### How long backups are kept, and what deletion means\n\nEach backup ages out on its own rolling cycle, within the period given in the retention table of our\n[Privacy Policy](https://gauzy.co/privacy). Backups are not individually edited to remove one record — that\nwould defeat what a backup is for — so a deleted record persists in backup copies until that cycle\ncompletes. During that window the data is not used for anything, and **we do not restore a backup in order\nto bring back data somebody asked us to delete.** A backup is restored to recover a system, not to\nresurrect a record.\n\n### Your own copies\n\nWe maintain backups for our operational continuity and for the recovery obligations we owe you. **They are\nnot a substitute for your own copies**, and the Service provides export so that you can keep some. If your\nbusiness could not tolerate the loss of the data you keep with us, export it on a schedule that reflects\nthat.\n\n## Retention, deletion and export\n\n### Getting your data out\n\nYou can export your data while your account is open — through the product's own export function where it\nhas one, through the application programming interface, or by asking us. Exports are in structured,\nmachine-readable formats — commonly CSV and JSON, with files delivered in their original format. Your\nproduct's annex says which export routes exist for it.\n\nIf you need an export we do not offer as a self-service function, write to legal@ever.co and we will\nproduce it. **We do not charge for a reasonable export, and we do not make leaving harder than joining.**\n\n### After your agreement ends\n\nYour data remains available for export for **30 days** after the agreement ends. After that window we\ndelete it. The same period, and the reasoning behind it, is in the retention table of our\n[Privacy Policy](https://gauzy.co/privacy).\n\nIf you need longer, ask before the window closes rather than after it.\n\n### Deletion on request\n\nWhere you delete something inside the Service, or ask us to delete your workspace:\n\n- **Live systems are updated straight away, with one exception.** In Ever Gauzy, deleting a screenshot, a time slot or a time entry in the web application is a soft deletion: it is hidden at once, but the record, and for a screenshot its stored image and thumbnail, stays on the running service until it is permanently deleted, which we do on request.\n- **Backup copies are not individually edited.** A deleted record persists in backups until each copy ages\n out on its rolling cycle, within the period stated in the Privacy Policy. During that window the data is\n not used for anything, and we do not restore a backup to bring it back.\n- **Copies held by a provider we engage** are deleted on our instruction, under the Article 28 terms in our\n contract with them.\n\nWhere we can keep something useful without keeping anyone identifiable — aggregate usage counts, for\nexample — we anonymise rather than delete, and once anonymised it is no longer personal data.\n\n### Retention you control\n\nFor data inside your workspace we are your processor, and **you set the retention period, not us**. Where a\nproduct lets you configure how long activity data, captured media, logs or archived records are kept, that\nsetting is yours and its default is documented in the annex. Choosing a period that is lawful and\nproportionate in your jurisdiction is part of your controller responsibility — see the shared\nresponsibility section.\n\nOur own retention periods, for the data where we are the controller, are published in full in the\n[Privacy Policy](https://gauzy.co/privacy) as a table with actual numbers rather than the phrase \"as long as\nnecessary\".\n\n### Legal hold\n\nWhere we are required to preserve data for a legal claim, an investigation or a regulatory request, we keep\nit until that ends, including any appeal period, and we do not delete it in the meantime. That obligation\noverrides the periods above. It applies to us; it will sometimes apply to you as well.\n\n### Disposing of hardware\n\nBecause we own our infrastructure, we also retire our own drives. Storage is encrypted at rest, and a drive\nis wiped, cryptographically erased or physically destroyed before it leaves our control. **A retired drive\nis never sold, donated or returned intact.**\n\n## Personnel\n\nWe are a small, long-standing team, and the people who build the Service are the people who operate it.\nThat has a security upside — nobody is administering a system they do not understand — and a limit we set\nout in the access control section rather than hide.\n\n**Everyone with access to production systems or customer data:**\n\n- is bound by **written confidentiality obligations that continue after their engagement ends**, whether\n they are an employee or a contractor;\n- works on a **need-to-know basis**, with access scoped to their role and granted for a purpose rather than\n by default;\n- uses **named individual accounts with multi-factor authentication**, never a shared login;\n- is briefed on how we expect data to be handled as part of joining, and again when something material\n changes;\n- **loses that access when they leave or when their role changes**, with the credentials they held rotated\n as part of the same step.\n\nContractors and anyone working on our behalf are engaged on terms at least as strict as those above. We do\nnot give a provider or a contractor access to customer data because it is convenient.\n\n### Devices\n\nMachines used to access production systems or customer data run full-disk encryption, lock automatically,\nare kept patched, and are protected by multi-factor authentication on the accounts they hold. A lost or\nstolen device is treated as an incident: access is revoked and credentials are rotated before anything else\nis decided.\n\n### Two things we do not claim\n\n- **We do not run formal background screening as a standard step**, and we are not going to describe an\n informal hiring process as vetting. Where your own contract or regulatory position requires screening for\n named individuals working on your account, tell us at legal@ever.co and we will arrange it.\n- **We do not run a certified annual security awareness training programme.** Security expectations are\n part of onboarding and part of how we review changes, but that is a practice, not a curriculum, and the\n two should not be confused on a page like this.\n\n## Providers and vendor management\n\n### The first question is whether we need one at all\n\nWe run our own infrastructure, and that is a security decision as much as a cost one: **a provider we do\nnot engage is a provider that cannot be breached on our behalf.** Before we take on a new one we ask\nwhether the function can be served by something we already operate. Often it can, and the answer is no\nprovider.\n\n### Before we engage one\n\n- **What data would it see**, and can we send it less — or none — and still get the benefit?\n- **What is its security posture**: its own certifications and reports, its published architecture, its\n breach history, how it handles vulnerability reports.\n- **Where does it process**, which establishment contracts with us, and which transfer mechanism applies if\n processing happens outside the European Economic Area.\n- **Will it sign a data processing agreement** meeting Article 28 GDPR, including the Standard Contractual\n Clauses where they are needed? If it will not, we do not engage it for anything involving personal data.\n- **How do we get out** — can we export what it holds, and can we replace it without rebuilding the\n product around its absence?\n\n### What the contract has to contain\n\nProcessing only on our documented instructions; confidentiality binding its personnel; security measures at\nleast as strict as our own; controls on its own sub-processors; assistance with data subject requests and\nwith security, breach notification and impact assessments; notification of a breach without undue delay;\ndeletion or return of the data when the engagement ends; and our right to the information needed to\ndemonstrate compliance.\n\n**A provider may not use your data for its own purposes.** Not to improve its own products, not to train\nmodels, not to build advertising profiles.\n\n### While it is engaged\n\nWe review the providers we use periodically, watch for security incidents and for changes in the legal\nbasis for transfers to them, and remove one when the feature it served is gone. A provider that is no\nlonger used does not stay on the list as decoration, and a provider that has been quietly replaced is a\nchange we announce.\n\n### How this appears to you\n\nEvery provider that processes personal data for the hosted Service is published on our\n[sub-processor page](https://gauzy.co/subprocessors), in three tiers: those always engaged, those engaged\nonly when you enable a specific feature, and those that appear only in a deployment you run yourself. That\npage also gives the advance notice period before we add or replace one, and the procedure for objecting.\n\n## Responsible disclosure\n\nIf you have found a security vulnerability in the Service, we want to hear about it, and we would rather\nhear about it from you than from someone using it.\n\n### How to report\n\nSend it to **abuse@ever.co**. If personal data is exposed, copy privacy@ever.co.\n\n**We do not operate a separate security mailbox, and we are not going to publish one we do not monitor.**\nThat address reaches the people who operate the Service.\n\nA useful report contains: what the issue is, the exact URL, endpoint or component, the steps to reproduce\nit, what an attacker could actually do with it, and anything you need from us to prove it. Screenshots or a\nshort recording help. Write in English.\n\n### What we do with it\n\n- **We acknowledge your report within three business days.**\n- We assess and reproduce it, and tell you what we found — including if we conclude it is not a\n vulnerability, and why.\n- We keep you updated while we fix it, and tell you when the fix is live.\n- We are happy to **credit you publicly** when the fix has shipped, or to leave you out of it. Your choice.\n\nSerious issues jump the queue. We fix a critical vulnerability as quickly as we can build and verify a fix,\nand we will tell you what the timeline looks like rather than leaving you guessing.\n\n### Our commitment to you\n\n**If you follow the rules below, we will not pursue legal action against you, we will not report you to\nanyone, and we will not ask your employer to.** We will treat your research as authorised access for the\npurposes of computer misuse legislation and our own [Terms of Service](https://gauzy.co/tos), and we will\ntreat your report as a contribution rather than a threat.\n\nThis protection covers you personally. It cannot cover you against a third party — if your testing reaches\none of our providers' systems, that is between you and them.\n\n### The rules\n\n- **Do not access, modify, delete or download anyone else's data.** Use your own test account. If you can\n reach another customer's data, that is the finding — stop there and tell us. You do not need to collect\n it to prove it.\n- **Stop as soon as you have demonstrated the issue.** Do not pivot, escalate further, or establish\n persistence.\n- **No denial of service**, volumetric testing, load testing or resource exhaustion.\n- **No social engineering** of our people, our customers or our providers, and no physical attempts against\n our premises or equipment.\n- **No spam, no mass account creation, no brute forcing** of live accounts.\n- **Give us a reasonable time to fix it before you publish** — 90 days from your report, or the day the fix\n ships, whichever comes first. If a fix is taking longer, talk to us; we will not use the clock to bury a\n finding.\n- **Obey the law**, and do not do anything that would harm the Service or the people who use it.\n\nReports produced by running an automated scanner against the live Service are not in scope, and running one\nis itself a breach of the testing section above.\n\n### Usually out of scope\n\nMissing security headers or best-practice settings with no demonstrated impact; self-inflicted\ncross-site scripting; missing rate limits without a working attack; outdated version banners without a\nworking exploit; results from a vulnerability scanner submitted without validation; social engineering;\nand issues in a third-party service rather than ours — report those to the provider that runs it.\n\nSend it anyway if you are unsure. We would rather triage a duplicate than miss a real one.\n\n### There is no paid bounty\n\n**We do not run a paid bug bounty programme and we do not pay for vulnerability reports.** We would rather\nsay that plainly at the start than have you invest a weekend on the assumption that we do. What you get is\na fast, straight response from the people who can fix it, and public credit if you want it.\n\n### Security issues in our open-source code\n\nWhere our software is published as open source, report a security issue in the code to the same address, in\nprivate. **Please do not open a public issue or a pull request that describes the vulnerability** — that\npublishes a working exploit to everyone running the code, including people who cannot patch quickly.\n\n## Shared responsibility — who secures what\n\nSecurity of a hosted service is split between the provider and the customer, and **most real incidents\nhappen on the customer's side of that line**: a shared password, a departed employee who kept access, an\nintegration token that outlived its purpose, a laptop left on a train. That is not a criticism. It is the\nreason this section exists and is specific.\n\n### What we are responsible for\n\n- The physical hardware, the facilities and the virtualisation platform.\n- The container platform, the operating systems and the runtime the application runs on.\n- Patching and updating everything in the list above, and the application itself.\n- Encryption in transit and at rest, and the management of our keys.\n- Network security, the edge, and keeping the origin off the public internet.\n- Availability, redundancy, backups of the Service, and restoring it after a failure.\n- Isolation between customers.\n- The security of our own code, our build pipeline and our deployment process.\n- Our own staff: their access, their devices, their confidentiality obligations.\n- The providers we engage, and the contracts that bind them.\n- Detecting, responding to and notifying you of incidents in our part.\n\n### What you are responsible for\n\n**Who gets in, and who stays in.** Your users, their roles and their permissions. Inviting people, removing\nthem the day they leave, reviewing periodically who still has access, and not sharing accounts. Choosing\nstrong, unique passwords and turning on multi-factor authentication where the product offers it. If your\norganisation uses single sign-on, the security of your identity provider is yours.\n\n**How you configure the product.** Every switch we give you is a decision you are making. Defaults are\nchosen to be safe, not to be right for you, and changing one is a deliberate act with consequences you own.\n\n**Your monitoring configuration.** For products that can capture workforce activity, **you decide which\ncapture features are switched on, how often they run, and who can see the result.** We supply the controls;\nwhich of them is lawful and proportionate in your organisation, in your jurisdiction, is your call. The\nnext block is about that in more detail, because it is the sharpest edge in this document.\n\n**Your retention settings.** Where the product lets you set how long data is kept, that period is yours.\nKeeping data longer than you need it is a risk you are choosing to hold, and it is one of the first things\na supervisory authority asks about.\n\n**Your integrations and your tokens.** Every third-party service you connect, every API key you issue,\nevery webhook you point somewhere and every access token you grant to an application or an automated agent.\nApproving an integration is you deciding to send your data somewhere else. Scope those grants narrowly,\nreview them, and revoke them when a person leaves or a project ends — a token issued by someone who has\nleft your organisation still works until you revoke it.\n\n**Your devices and your endpoints.** The laptops, phones and desktops your people use to reach the Service,\nincluding any desktop or mobile application we publish. Full-disk encryption, screen locks, patching,\nmalware protection and what happens when a device is lost are yours. We cannot protect an account from a\ncompromised machine that is already signed into it.\n\n**Your data, and what you put in it.** What you choose to upload, whether it belongs in this product at\nall, and whether you have the right to process it. Also your own copies — see the backups section.\n\n**Telling your own people.** Your privacy notice to your staff, customers and site visitors is yours to\nwrite. Ours describes what we do; it is not your notice, and our cookie banner is not your consent\nmechanism.\n\n### Where you use the Service to monitor people\n\nSome of our products capture information about how an organisation's own personnel work — which can include\nscreenshots, application and website usage, activity rates derived from keyboard and mouse events, and,\nwhere you switch them on, webcam stills, audio or screen recordings.\n\nFor all of that data, **you are the controller and we are your processor.** You decide why it is collected\nand how; we process it on your instructions. That split has consequences you should read carefully:\n\n- **The lawful basis is yours to establish.** Consent is generally not a valid basis in an employment\n relationship because of the imbalance of power between employer and employee, so a signed form is\n unlikely to be the answer you need.\n- **The notice to the monitored person is yours to give**, in advance, covering each type of capture that\n is switched on.\n- **The data protection impact assessment is yours to complete** where one is required, which it commonly\n is for systematic monitoring of employees. We will help: ask privacy@ever.co for the information you\n need about data categories, capture frequencies, retention, storage locations, sub-processors and\n transfer mechanisms, and we will provide it.\n- **Consultation with a works council, trade union or employee representative body is yours to run** where\n local law requires it — which it does in several European countries, before the system is introduced, not\n after.\n- **Some capture types are restricted or prohibited in some countries.** Continuous screen capture in\n particular is treated very differently across Europe.\n\n**We make no claim that any monitoring feature is lawful for you to use, anywhere.** A feature existing in\nthe product is not advice that you may switch it on, and our providing it is not our assessment of your\nlegal position. Where you enable one, you confirm to us that you have done the work above — the Data\nProcessing Addendum and the [Terms of Service](https://gauzy.co/tos) put that in binding terms, including\nwho answers for it if it turns out not to have been done.\n\n### Where you run the software yourself\n\nIf you deploy our open-source software on your own infrastructure, **everything above is yours** — the\nservers, the database, the storage, the backups, the encryption, the patching, the access control and the\nincident response. You are the controller of the data in it. We have no access to it, we are not your\nprocessor for it, and no commitment on this page applies to it.\n\n## How to reach us about security\n\n- **A vulnerability you have found** — abuse@ever.co. Read the responsible disclosure section first; it\n tells you what protection you have and what rules go with it.\n- **An incident, a compromised account, or something you can see that you should not be able to see** —\n abuse@ever.co, immediately. Copy privacy@ever.co if personal data is involved.\n- **A security questionnaire, a due diligence request, a request for the Data Processing Addendum, evidence\n of a restore test, or permission to test the Service** — legal@ever.co.\n- **A question about personal data, a data subject request, or a breach affecting you** —\n privacy@ever.co.\n\nAll of these addresses are monitored by the people who operate the Service. By post: **Ever Technologies LTD**,\nMladost 2, bl. 211, ent. A, Sofia 1799, Bulgaria — registered in Bulgaria under company number 204599535. We\ncorrespond in English.\n\nYou may also complain to a data protection supervisory authority. Ours is the\nCommission for Personal Data Protection (Комисия за защита на личните данни), the CPDP, at\n[https://www.cpdp.bg/](https://www.cpdp.bg/). You may instead complain\nto the authority for the country where you live or work.\n\nThis document is version 1.0.3 of the Security Overview for\n[gauzy.co](https://gauzy.co), in force from 2026-10-24. It describes the measures in\nplace at that date. Earlier versions, with the dates they applied, are at\n[https://gauzy.co/security](https://gauzy.co/security)."
379
+ "markdown": "## Security overview\r\n\r\nThis page describes how we protect Ever Gauzy Cloud and the data you keep in it. It is written to be\r\nchecked rather than admired: every measure below is something we actually run, and where we do not do\r\nsomething that a page like this usually claims, **we say so instead of leaving a gap you have to guess at**.\r\n\r\nThe short version:\r\n\r\n- We run the Service on **hardware we own and operate ourselves**, not as a tenant in someone else's cloud.\r\n The database, the object storage and the secrets store are ours. That is why the list of providers with\r\n access to your data is short.\r\n- Traffic is encrypted in transit, storage is encrypted at rest, and off-site backup copies are encrypted\r\n **before they leave our network**, so the provider holding them holds ciphertext.\r\n- Production changes go through version control and review. Hand edits on live systems are the exception,\r\n not the routine.\r\n- **We hold no security certification, and we do not claim one.** The next section says exactly what we do\r\n instead.\r\n\r\n### What this page covers\r\n\r\nIt covers the hosted Service we operate at [gauzy.co](https://gauzy.co) — the application, its\r\ninterfaces, and the infrastructure underneath them.\r\n\r\nIt does **not** cover two things that are easy to confuse with it:\r\n\r\n- **A copy of the software you run yourself.** Where our software is published as open source and you deploy\r\n it on your own infrastructure, you are the operator. You choose its hosting, its storage and every other\r\n component, we have no access to it, and nothing on this page describes it.\r\n- **Your own systems and your own configuration.** Which of your people have access, how you configure the\r\n product, what you connect it to and how you secure the devices your team uses are yours. The shared\r\n responsibility section near the end of this page sets out that split in detail, because most real\r\n incidents happen on one side of that line rather than in the middle of it.\r\n\r\n### What this page is, legally\r\n\r\nThis is a factual description of the measures in place at the version and date shown at the end. It is\r\n**not a warranty, and it does not create a service level commitment.** Our contractual security obligations\r\nare in our Data Processing Addendum and in the [Terms of Service](https://gauzy.co/tos), and those are the\r\ndocuments that bind us.\r\n\r\nWe change these measures as the systems and the threats change. **We will not make a change that materially\r\nreduces the protection described here**, and when the description changes we republish this page with a new\r\nversion and date.\r\n\r\n### One thing we will not tell you\r\n\r\n**No system is perfectly secure, and a vendor who tells you otherwise is selling something.** Everything\r\nbelow reduces risk; none of it eliminates it. What we can commit to is that we take these measures\r\nseriously, that we test the ones that can be tested, and that when something goes wrong we tell you rather\r\nthan manage the news.\n\n## Certifications — what we hold, and what we do not\r\n\r\n**We do not hold a SOC 2 report of any type. We are not certified to ISO/IEC 27001. We hold no other\r\nsecurity certification or attestation, and we do not claim one.**\r\n\r\nWe are stating that first, in plain words, because a security page that leaves the question open is doing\r\nso on purpose. If a proposal, a sales conversation, a questionnaire response, a directory listing or a\r\nthird-party summary has told you that we are SOC 2 or ISO 27001 certified, **it is wrong** — please send it\r\nto legal@ever.co so we can get it corrected at the source.\r\n\r\n### What we do instead\r\n\r\nWe use two published control sets as a working checklist rather than as a badge:\r\n\r\n- **The control areas of ISO/IEC 27001 Annex A** — access control, cryptography, operations security,\r\n supplier relationships, incident management, continuity — as the structure for reviewing our own\r\n practices. This page is organised broadly along those lines.\r\n- **Article 32 GDPR**, which is not optional for us and is the standard our customers can actually hold us\r\n to: technical and organisational measures appropriate to the risk, kept under review as the risk changes.\r\n\r\n**No external auditor has examined either.** These are self-assessments, and you should read them as such.\r\nAn unaudited framework alignment and a certification are different things, and we will not blur the two by\r\nwriting \"aligned with\" in a way that reads like \"certified to\".\r\n\r\n### What you can have instead of a certificate\r\n\r\nA certificate is a proxy for evidence. Where we cannot give you the proxy, we will give you the evidence:\r\n\r\n- **This page**, which is specific enough to be checked against what we actually do.\r\n- **The security measures schedule in our Data Processing Addendum**, which is contractual rather than\r\n descriptive — if a measure matters to you, that is where to pin it down.\r\n- **A completed security questionnaire.** Send yours to legal@ever.co. We answer them, we answer them\r\n honestly, and where the honest answer is \"no\" you will get \"no\" rather than a paragraph that avoids the\r\n question.\r\n- **Evidence on request** for specific points — a description of the architecture, the outcome of our most\r\n recent restore test, the current sub-processor list, the transfer mechanism for a named provider.\r\n- **A right to audit**, on the terms set out in the Data Processing Addendum.\r\n\r\n### If that changes\r\n\r\nIf we obtain a certification or an independent report, this page will say which one, who issued it, what\r\nits scope covers and what it excludes — because the scope is the part that matters and the part most often\r\nleft out. Until that day, the sentence at the top of this section is the whole answer.\n\n## Encryption\r\n\r\n### In transit\r\n\r\nAll traffic between you and the Service travels over TLS. HTTPS is enforced and plain HTTP requests are\r\nredirected rather than served, so a request that arrives unencrypted never reaches the application with its\r\ncontents intact. We keep the accepted protocol versions and cipher suites current and retire weak ones as\r\nthe ecosystem does.\r\n\r\nTraffic between our own sites, and between our infrastructure and the edge network in front of it, runs\r\nover encrypted connections. Traffic to the providers we engage runs over TLS to their published endpoints.\r\n\r\n### At rest\r\n\r\n- **Storage holding customer content, databases and backups is encrypted at rest.** That protects against\r\n the loss, theft or improper disposal of physical media — it does not protect against a compromise of a\r\n running system that already has the data decrypted, and we would rather you knew which threat it answers.\r\n- **Off-site backup copies are encrypted before they leave our network**, with keys we hold and manage\r\n ourselves. The provider storing those copies holds ciphertext and nothing else: it cannot read them, and\r\n neither can anyone who obtains them from it.\r\n\r\n### At the application layer\r\n\r\nSome values are too sensitive to rely on storage-level encryption alone, so they are encrypted again inside\r\nthe application before they are written:\r\n\r\n- **Credentials, access tokens and API keys** — including tokens you grant us for a third-party integration.\r\n- **Secrets used to operate the Service** — held in a secrets manager we run ourselves, injected into\r\n workloads at runtime, and **never committed to source control**.\r\n\r\n**Passwords are never stored in a form we can read.** They are put through a modern one-way hashing\r\nfunction with a per-password salt. We cannot recover your password, tell you what it is, or send it back to\r\nyou — a reset is the only route, and any service that offers you the alternative is storing it wrongly.\r\n\r\nWhere a specific product stores a particular credential differently from the description above, its annex\r\nto this page says so and says why. **We would rather publish an awkward exception than a tidy sentence that\r\nis not true everywhere.**\r\n\r\n### Keys\r\n\r\nEncryption keys are held in our self-hosted secrets manager, separately from the data they protect, and are\r\naccessible only to the systems and people that need them. Keys are rotated when a person with access leaves,\r\nwhen we have reason to think a key has been exposed, and on a periodic schedule. **The keys for off-site\r\nbackups are held independently of the infrastructure they protect**, so that losing a site does not also\r\nlose the ability to decrypt its backups.\n\n## Access control\r\n\r\n### Your access to your own data\r\n\r\nAccess inside your workspace is yours to configure. The Service gives you named user accounts, roles and\r\npermissions, workspace administrators who can invite and remove people, and session controls. Where a\r\nproduct supports single sign-on, multi-factor authentication or IP restrictions, its annex says so and its\r\ndocumentation says how to switch them on.\r\n\r\n**Turning those controls on, and reviewing who still needs access, is your job — see the shared\r\nresponsibility section.** The most common way an account is compromised is not a flaw in our systems; it is\r\na shared password, a former employee who was never removed, or an integration token that outlived its\r\npurpose.\r\n\r\n### Our access to production systems\r\n\r\n- **Named individual accounts, never shared logins.** Every action in production is attributable to a\r\n person.\r\n- **Multi-factor authentication** on the systems and administrative accounts that support it.\r\n- **Least privilege.** Access is scoped to the narrowest role that lets the work happen, and elevated\r\n access is granted for a purpose rather than held permanently by default.\r\n- **Changes go through Git.** The configuration of the live environment is held in version control and\r\n applied automatically, so a change is reviewable, attributable and reversible. Direct hands-on changes to\r\n a running system are an exception for incidents and hands-on maintenance, not the daily routine — and the\r\n automation reverts undeclared edits rather than leaving them in place.\r\n- **Removal on departure or role change.** Access is revoked when someone leaves or no longer needs it, and\r\n credentials they held are rotated. Access is reviewed periodically as well, because revocation lists rot.\r\n\r\n### Access to your content\r\n\r\nWe access the content in your workspace only where it is necessary to operate, secure or support the\r\nService — for example to investigate a fault you have reported, or to respond to a security incident. **We\r\ndo not browse customer content, and we do not use it to build products, train models or produce insights\r\nfor anyone else.**\r\n\r\nSupport access is granted for the task in hand and logged. Where a product offers a mechanism for you to\r\ngrant or revoke support access yourself, its annex describes it.\r\n\r\n### Being honest about the limits\r\n\r\n**We are a small team, and several of us hold administrative access because operating this infrastructure\r\nrequires it.** A large organisation would separate the person who approves a change from the person who\r\nmakes it; at our size that separation is partial, and we are not going to describe it as complete. What we\r\ndo instead is make every change visible: production state comes from version control, actions are\r\nattributable to named accounts, and administrative events are logged and monitored.\n\n## Infrastructure and hosting\r\n\r\n### We run our own infrastructure\r\n\r\nThe Service does not run in a rented public cloud account. It runs on **physical servers we own, in\r\nfacilities we control, inside the European Union**, on a virtualisation and container platform we operate\r\nourselves. The database, the object storage, the message and cache layers, the secrets store and the\r\ndeployment system are all components we run — not managed services bought from a third party.\r\n\r\nThat is an unusual answer in this market, and it has a direct consequence you can verify: **the list of\r\nproviders with access to your data is short.** Most of what would otherwise be a sub-processor is simply a\r\nservice we operate.\r\n\r\nIn outline, and deliberately without the detail that would help someone attack it:\r\n\r\n- **Hardware and virtualisation we own**, clustered so that losing a machine does not lose a service.\r\n- **A container orchestration platform** running the applications, with the control plane replicated across\r\n multiple machines.\r\n- **PostgreSQL**, run by an operator that handles replication, failover and continuous archiving of the\r\n write-ahead log.\r\n- **S3-compatible object storage** we host, for files and media, with data replicated across nodes.\r\n- **A secrets manager we host**, from which credentials are injected into workloads at runtime.\r\n- **Declarative configuration held in Git** and applied automatically, so the live environment matches a\r\n reviewed commit rather than an accumulation of manual changes.\r\n\r\nA content delivery and security network sits in front of all of it, and our source code, build pipeline and\r\ncontainer registry are hosted with a third-party provider. Those are genuine third parties even though the\r\ninfrastructure behind them is ours, and they are named on our\r\n[sub-processor page](https://gauzy.co/subprocessors).\r\n\r\n### Redundancy and failure domains\r\n\r\nServices that can run on more than one machine run on more than one machine. Databases are replicated,\r\nobject storage is replicated across nodes, and the systems that hold backups run on **a physically separate\r\ncluster from the systems they protect** — so a failure that takes out the live environment does not take\r\nits backups with it.\r\n\r\n**This is not the same as multi-region redundancy, and we will not describe it as such.** We can lose\r\nmachines without losing data or availability. A whole-site failure means recovering from off-site copies,\r\nwhich is a matter of hours rather than seconds. The continuity section is specific about what that means.\r\n\r\n### Separation between customers\r\n\r\nIsolation between customers is **logical, not physical**, unless you have separately contracted for a\r\ndedicated deployment. Depending on the product that means a separate database per customer, a separate\r\nnamespace and deployment per customer site, or tenant scoping enforced in the application and the data\r\nmodel. **Your product's annex states which model applies to you** — it is a question worth asking of any\r\nvendor, and worth a straight answer.\r\n\r\n### Separation between environments\r\n\r\nDevelopment, staging and production are separate environments with separate credentials and separate data\r\nstores. Production credentials are not available in development or staging, and **we do not copy production\r\ndata into a development environment**. Test data is synthetic or anonymised.\r\n\r\n### Third-party managed services, where they are used\r\n\r\nA small number of products use a third-party managed database, managed storage or hosted platform for a\r\nspecific function rather than our own. Where that is the case it is named in that product's annex and it\r\nappears in the always-engaged tier of the sub-processor list. **We do not describe the estate as fully\r\nself-hosted where a product's data actually sits somewhere else.**\n\n## Network security\r\n\r\n### At the edge\r\n\r\nEvery request to the Service passes through a content delivery and security network before it reaches us.\r\nThat layer terminates TLS, absorbs volumetric denial-of-service traffic, applies a web application firewall\r\nand filters automated abuse. It also means the address of our origin infrastructure is not the address your\r\nbrowser connects to.\r\n\r\n### The origin is not exposed\r\n\r\n**Our application servers do not publish inbound ports to the public internet.** Traffic reaches them\r\nthrough outbound-initiated tunnels established from inside our network to the edge, so there is no\r\npublic listener to scan, and no direct path from the internet to an origin host.\r\n\r\n### Inside the network\r\n\r\n- Databases, object storage, the secrets manager and internal services sit on **segmented private networks\r\n and are not reachable from the internet**. They accept connections only from the workloads that need them.\r\n- **Administrative interfaces are not published publicly.** Reaching them requires being on the internal\r\n network, through a controlled route with authenticated access.\r\n- Outbound connectivity from workloads is limited to what the workload actually needs, so a compromised\r\n component has fewer places to send data and fewer places to fetch a second stage from.\r\n\r\n### Abuse and automation\r\n\r\nSign-up, sign-in, password reset and contact endpoints are protected against automated abuse by rate\r\nlimiting and by a bot-detection challenge. **That challenge is strictly necessary to keep the Service\r\nworking and available** — it protects accounts from credential-stuffing and forms from bulk abuse — which\r\nis why it runs without a consent prompt, as explained in our [Cookie Policy](https://gauzy.co/cookies). It\r\nis not used for analytics, profiling or advertising.\r\n\r\n### What we monitor for\r\n\r\nUnusual authentication patterns, unexpected administrative activity, sudden changes in traffic shape and\r\nfailures of the components above raise alerts that a person sees. The logging and monitoring section\r\ndescribes what is recorded and for how long.\n\n## Application security\r\n\r\n### How a change reaches production\r\n\r\nEvery change to the Service starts as a change to source code in version control and follows the same\r\nroute:\r\n\r\n1. **A pull request**, reviewed by someone other than its author on the branches that matter.\r\n2. **Automated checks** — build, tests, static analysis and dependency scanning — which must pass.\r\n3. **Promotion through environments.** Changes move development to staging to production, in that order.\r\n Production is deployed from the release branch and from nothing else.\r\n4. **A container image built by the pipeline** from a reviewed commit, identified by its digest, deployed\r\n by the same automation that holds the environment's configuration in Git.\r\n\r\n**Nobody edits production by hand as a way of shipping a change.** The deployment system reconciles the\r\nlive environment against the version-controlled definition, so an undeclared edit is reverted rather than\r\nquietly kept.\r\n\r\n### Automated checks on every change\r\n\r\n- **Dependency scanning**, with alerts for known vulnerabilities in the packages we use, and updates\r\n applied on a defined schedule with a faster path for serious issues.\r\n- **Static analysis** of our source code, run on each change, flagging the classes of defect it can find\r\n before the code ships.\r\n- **Secret scanning** across our repositories, with pushes blocked where a credential is detected. Where a\r\n credential has been exposed, we treat rotating it at the provider as the fix — deleting it from the code\r\n is not the fix.\r\n- **Automated tests**, which have to pass before a change can be merged or released.\r\n\r\nDependencies are installed through an internal package cache we run, which gives us a consistent, auditable\r\nview of what actually went into a build.\r\n\r\n### In the application itself\r\n\r\n- Database access goes through parameterised queries and a mapping layer rather than assembled strings.\r\n- Output is encoded for its context, and state-changing requests are protected against cross-site request\r\n forgery.\r\n- Authentication, session handling and password storage use established libraries rather than anything we\r\n invented; passwords are hashed as described in the encryption section.\r\n- Authorisation is enforced on the server for every request. A hidden control in the interface is not an\r\n access control.\r\n- Input is validated at the boundary, and uploads are constrained by type and size.\r\n- Browser-facing responses carry security headers, including a content security policy on the surfaces\r\n where one can be applied without breaking the product.\r\n\r\n### The honest limit\r\n\r\n**Automated tooling finds categories of defect, not all defects.** Review catches more, and neither catches\r\neverything. That is why the sections on logging, incident response and responsible disclosure exist: they\r\nare what we rely on for the bugs that get through, and something always gets through.\n\n## Testing\r\n\r\n### What we test, and how often\r\n\r\n- **Automated test suites** run on every change, in the pipeline, before anything can be merged or\r\n released.\r\n- **Static analysis and dependency scanning** run on every change as well, as described in the previous\r\n section.\r\n- **Restore tests.** We restore backups into a scratch environment and check the result against the source\r\n rather than assuming a backup that completed is a backup that works. This is the test most often skipped\r\n in the industry and the one most often needed.\r\n- **Manual review** of changes that touch authentication, authorisation, payment flows, data export or the\r\n handling of credentials — the areas where an automated check is least likely to notice that the logic is\r\n wrong.\r\n- **Configuration review** of the version-controlled definition of the live environment, which is where a\r\n surprising share of real exposure lives.\r\n\r\n### Penetration testing — what we do not have\r\n\r\n**We have not commissioned an independent penetration test of the hosted Service, and we do not claim one.\r\nThere is no test report for us to send you, and any summary suggesting otherwise is not ours.**\r\n\r\nWe would rather write that sentence than the usual formulation about \"regular security assessments\" that\r\nturns out on inspection to mean an automated scanner. When we do commission a test, this page will say who\r\nperformed it, when, what was in scope and what was excluded.\r\n\r\n### Testing the Service yourself\r\n\r\n**You may not run security testing against the Service without our written agreement** — automated\r\nscanning, load testing, fuzzing and exploitation attempts included. From our side an unannounced scan is\r\nindistinguishable from an attack, and it will be handled as one.\r\n\r\nIf you want to test, we will usually say yes. Write to legal@ever.co and we will agree the scope, the\r\nwindow, the accounts you use and how findings are reported. The conditions are the obvious ones: your own\r\ntest data only, no other customer's data touched, nothing that degrades the Service for anyone else, and\r\nfindings to us before they go anywhere else.\r\n\r\nIf you have already found something without meaning to, do not test further — go straight to the\r\nresponsible disclosure section, which is written for exactly that situation and offers you protection for\r\nit.\n\n## Logging and monitoring\r\n\r\n### What we record\r\n\r\n- **Authentication events** — sign-in, sign-out, failed attempts, password and multi-factor changes.\r\n- **Administrative and security-relevant events** — permission changes, invitations and removals, token\r\n creation and revocation, changes to workspace settings, access to production systems.\r\n- **Application programming interface access** — the request, its source, its outcome and its timing.\r\n- **Infrastructure and platform events** — deployments, configuration changes, node and service state.\r\n- **Application errors and diagnostics** — enough to work out what broke.\r\n\r\n### What we try not to record\r\n\r\nWe do not deliberately log passwords, session tokens, API keys or payment card numbers, and where such a\r\nvalue would otherwise appear in a log or an error report it is redacted. **We will not claim this is\r\nperfect** — a value can end up in a stack trace nobody predicted. When we find one, we treat it as an\r\nexposure: the credential is rotated and the log entry removed.\r\n\r\nWhere a product sends error reports to a third-party error-tracking service, that service is named in the\r\nproduct's annex and on the [sub-processor page](https://gauzy.co/subprocessors), because an error report can\r\ncarry an identifier for the person who hit the error.\r\n\r\n### How long we keep it\r\n\r\n| Log type | Retention |\r\n|---|---|\r\n| Security, access and audit logs | **12 months** |\r\n| Application and error logs, crash reports, diagnostics | **90 days** |\r\n\r\nThese are the same periods published in the retention table of our\r\n[Privacy Policy](https://gauzy.co/privacy), and if the two ever disagree the Privacy Policy governs.\r\n\r\nAccess to logs is restricted to the people whose job needs it, and log access is itself an administrative\r\nevent that gets recorded.\r\n\r\n### Monitoring and alerting\r\n\r\nMetrics from the platform, the databases, the storage layer and the applications feed a monitoring system\r\nthat raises alerts when something is wrong — a service failing, a certificate close to expiry, a backup\r\nthat did not complete, a disk filling, an unusual pattern of authentication failures. **Alerts go to people,\r\nnot to a dashboard nobody opens.**\r\n\r\nWe also run an **independent external check** that watches the Service and the monitoring system itself\r\nfrom outside our network, and raises an alarm if either goes quiet. Monitoring that cannot survive the\r\nfailure it is meant to detect is decoration, and a failed backup that alerts nobody is the standard way a\r\nrecovery turns out to be impossible.\n\n## Incident response\r\n\r\n### How we handle an incident\r\n\r\n1. **Detect and record.** An alert, a customer report or a disclosure from a researcher opens an incident,\r\n with one named person responsible for running it.\r\n2. **Assess and classify.** What happened, what is affected, whether personal data is involved, whose data,\r\n and how bad it could get. Severity drives everything that follows.\r\n3. **Contain.** Cut the access, revoke the credential, isolate the component, block the route — stopping it\r\n comes before understanding it.\r\n4. **Eradicate and recover.** Remove the cause, restore service, restore data from backup where that is\r\n what is needed, and verify the result rather than assuming it.\r\n5. **Notify.** Under the timescales below.\r\n6. **Review afterwards.** A written review of what happened, why, what we changed, and what would have\r\n caught it sooner.\r\n\r\n### Telling you about it\r\n\r\n**Where we are the controller** — for the data described in our [Privacy Policy](https://gauzy.co/privacy) as\r\nours — and there is a personal data breach that is likely to result in a risk to people:\r\n\r\n- We notify the CPDP **within 72 hours of becoming aware of it**, as\r\n Article 33 GDPR requires, and where we cannot give complete information in that window we give what we\r\n have and follow up.\r\n- We notify the affected people **without undue delay** where the breach is likely to result in a high risk\r\n to them.\r\n\r\n**Where we are your processor** — everything inside your workspace, including data your organisation\r\ncollects about its own personnel — you are the controller and the notification duty to the authority is\r\nyours, so ours is to you:\r\n\r\n- We notify you **without undue delay after we become aware, and in any event within 72 hours.**\r\n- We give you what Article 33(3) requires you to have: the nature of the breach, the categories and\r\n approximate number of people and records involved, the likely consequences, the measures we have taken or\r\n propose, and a contact point who can answer follow-up questions.\r\n- We assist you with your own notifications and with any questions from your supervisory authority. Our\r\n Data Processing Addendum sets that out contractually.\r\n\r\nWe notify you at the administrative contact addresses on your account. **Keeping those current is worth\r\nfive minutes of your time**, because a notification that reaches a mailbox nobody reads has been made and\r\nnot received.\r\n\r\n### What we will not do\r\n\r\n- **We will not sit on a breach** to work out how it looks, or wait until we have a complete picture before\r\n telling you something is wrong.\r\n- **We will not require you to sign anything** — a non-disclosure agreement, a waiver, an amended contract —\r\n as a condition of being told.\r\n- **We will not describe an incident as something else.** If data was accessed, the notice says data was\r\n accessed.\r\n\r\n### Reporting an incident to us\r\n\r\nIf you think an account has been compromised, credentials have leaked, or you have seen something in the\r\nService you should not be able to see, tell us immediately at abuse@ever.co. If personal data is\r\ninvolved, copy privacy@ever.co. Both addresses are monitored and reach the people who operate the\r\nService.\r\n\r\nIf you have found a vulnerability rather than an incident in progress, the responsible disclosure section\r\nbelow is the route, and it offers you protection that this section does not.\n\n## Continuity and backups\r\n\r\n### Three tiers, and why there are three\r\n\r\nBackups fail in three different ways — the copy is too old, the copy is on the machine that died, or the\r\ncopy is in the building that burned. So there are three tiers, each answering a different failure:\r\n\r\n- **Tier 1 — a fast local copy.** Recent database backups are staged on storage separate from the live\r\n database, on a different machine. This is the tier that answers \"somebody dropped the wrong table twenty\r\n minutes ago\" and it restores quickly because the data never left the site.\r\n- **Tier 2 — a durable on-site copy.** Everything is copied into a **private, backup-only object store that\r\n runs on a physically separate cluster from the systems it protects**, with its own credentials, reachable\r\n only from inside our network and never from the internet. Losing the cluster that runs the Service does\r\n not lose the cluster that holds its backups.\r\n- **Tier 3 — an off-site copy with a third party.** A further copy goes to object storage operated by an\r\n external provider, in a different place entirely, **encrypted before it leaves our network**. The\r\n provider stores ciphertext and holds no key. This is the tier that answers the loss of the whole site.\r\n\r\nDeliberately, **the credentials that run the Service are not the credentials that can delete its backups.**\r\n\r\n### What is backed up\r\n\r\n- **Databases** — continuous archiving of the write-ahead log for point-in-time recovery, plus periodic\r\n full backups and separate logical dumps. Two mechanisms, because they fail differently.\r\n- **Object storage** — the files and media held for you, mirrored to the backup tiers.\r\n- **Cluster and platform state**, including the configuration needed to rebuild the control plane.\r\n- **The secrets store**, snapshotted on a short cycle.\r\n- **The version-controlled configuration** that defines the environment, which is what makes a rebuild\r\n possible rather than merely theoretical.\r\n\r\n**Not backed up, on purpose:** caches and queues, which are disposable by design and reconstructed on\r\nrestart, and build artefacts that the pipeline can regenerate. Backing those up would add cost and\r\nrestore time for nothing.\r\n\r\n### Recovery objectives\r\n\r\nThese are the objectives we design and test against. **They are internal targets, not contractual\r\ncommitments** — no service level applies unless we have separately agreed one with you in writing, as the\r\n[Terms of Service](https://gauzy.co/tos) says.\r\n\r\n| Scenario | Recovery point — how much data is at risk | Recovery time — how long it takes |\r\n|---|---|---|\r\n| Loss of a machine or a service | None. Replicated components carry on | Minutes, usually without anyone noticing |\r\n| Database corruption or an unwanted change | Minutes — continuous archiving supports point-in-time recovery | Hours, depending on the size of the database |\r\n| Loss of the storage holding files | Last completed replication cycle | Hours |\r\n| **Loss of the whole site** | Last completed off-site cycle | **Hours to days.** This is the honest number |\r\n\r\n**We are candid about the last row.** We are resilient to losing machines; we are not a multi-region\r\nservice that fails over silently, and we will not describe ourselves as one.\r\n\r\n### We test restores\r\n\r\nA backup nobody has restored is a guess. We restore into a scratch environment and verify the result\r\nagainst the source — record counts, integrity checks, and whether the restored system actually starts —\r\nrather than trusting a job that reported success. Restore drills are documented, and the outcome of the\r\nmost recent one is available to customers on request at legal@ever.co.\r\n\r\nMonitoring alerts on a backup that is late, failing or stale, because **a backup system that fails silently\r\nis functionally the same as no backup system**, and is discovered at the worst possible moment.\r\n\r\n### How long backups are kept, and what deletion means\r\n\r\nEach backup ages out on its own rolling cycle, within the period given in the retention table of our\r\n[Privacy Policy](https://gauzy.co/privacy). Backups are not individually edited to remove one record — that\r\nwould defeat what a backup is for — so a deleted record persists in backup copies until that cycle\r\ncompletes. During that window the data is not used for anything, and **we do not restore a backup in order\r\nto bring back data somebody asked us to delete.** A backup is restored to recover a system, not to\r\nresurrect a record.\r\n\r\n### Your own copies\r\n\r\nWe maintain backups for our operational continuity and for the recovery obligations we owe you. **They are\r\nnot a substitute for your own copies**, and the Service provides export so that you can keep some. If your\r\nbusiness could not tolerate the loss of the data you keep with us, export it on a schedule that reflects\r\nthat.\n\n## Retention, deletion and export\r\n\r\n### Getting your data out\r\n\r\nYou can export your data while your account is open — through the product's own export function where it\r\nhas one, through the application programming interface, or by asking us. Exports are in structured,\r\nmachine-readable formats — commonly CSV and JSON, with files delivered in their original format. Your\r\nproduct's annex says which export routes exist for it.\r\n\r\nIf you need an export we do not offer as a self-service function, write to legal@ever.co and we will\r\nproduce it. **We do not charge for a reasonable export, and we do not make leaving harder than joining.**\r\n\r\n### After your agreement ends\r\n\r\nYour data remains available for export for **30 days** after the agreement ends. After that window we\r\ndelete it. The same period, and the reasoning behind it, is in the retention table of our\r\n[Privacy Policy](https://gauzy.co/privacy).\r\n\r\nIf you need longer, ask before the window closes rather than after it.\r\n\r\n### Deletion on request\r\n\r\nWhere you delete something inside the Service, or ask us to delete your workspace:\r\n\r\n- **Live systems are updated straight away, with one exception.** In Ever Gauzy, deleting a screenshot, a time slot or a time entry in the web application is a soft deletion: it is hidden at once, but the record, and for a screenshot its stored image and thumbnail, stays on the running service until it is permanently deleted, which we do on request.\r\n- **Backup copies are not individually edited.** A deleted record persists in backups until each copy ages\r\n out on its rolling cycle, within the period stated in the Privacy Policy. During that window the data is\r\n not used for anything, and we do not restore a backup to bring it back.\r\n- **Copies held by a provider we engage** are deleted on our instruction, under the Article 28 terms in our\r\n contract with them.\r\n\r\nWhere we can keep something useful without keeping anyone identifiable — aggregate usage counts, for\r\nexample — we anonymise rather than delete, and once anonymised it is no longer personal data.\r\n\r\n### Retention you control\r\n\r\nFor data inside your workspace we are your processor, and **you set the retention period, not us**. Where a\r\nproduct lets you configure how long activity data, captured media, logs or archived records are kept, that\r\nsetting is yours and its default is documented in the annex. Choosing a period that is lawful and\r\nproportionate in your jurisdiction is part of your controller responsibility — see the shared\r\nresponsibility section.\r\n\r\nOur own retention periods, for the data where we are the controller, are published in full in the\r\n[Privacy Policy](https://gauzy.co/privacy) as a table with actual numbers rather than the phrase \"as long as\r\nnecessary\".\r\n\r\n### Legal hold\r\n\r\nWhere we are required to preserve data for a legal claim, an investigation or a regulatory request, we keep\r\nit until that ends, including any appeal period, and we do not delete it in the meantime. That obligation\r\noverrides the periods above. It applies to us; it will sometimes apply to you as well.\r\n\r\n### Disposing of hardware\r\n\r\nBecause we own our infrastructure, we also retire our own drives. Storage is encrypted at rest, and a drive\r\nis wiped, cryptographically erased or physically destroyed before it leaves our control. **A retired drive\r\nis never sold, donated or returned intact.**\n\n## Personnel\r\n\r\nWe are a small, long-standing team, and the people who build the Service are the people who operate it.\r\nThat has a security upside — nobody is administering a system they do not understand — and a limit we set\r\nout in the access control section rather than hide.\r\n\r\n**Everyone with access to production systems or customer data:**\r\n\r\n- is bound by **written confidentiality obligations that continue after their engagement ends**, whether\r\n they are an employee or a contractor;\r\n- works on a **need-to-know basis**, with access scoped to their role and granted for a purpose rather than\r\n by default;\r\n- uses **named individual accounts with multi-factor authentication**, never a shared login;\r\n- is briefed on how we expect data to be handled as part of joining, and again when something material\r\n changes;\r\n- **loses that access when they leave or when their role changes**, with the credentials they held rotated\r\n as part of the same step.\r\n\r\nContractors and anyone working on our behalf are engaged on terms at least as strict as those above. We do\r\nnot give a provider or a contractor access to customer data because it is convenient.\r\n\r\n### Devices\r\n\r\nMachines used to access production systems or customer data run full-disk encryption, lock automatically,\r\nare kept patched, and are protected by multi-factor authentication on the accounts they hold. A lost or\r\nstolen device is treated as an incident: access is revoked and credentials are rotated before anything else\r\nis decided.\r\n\r\n### Two things we do not claim\r\n\r\n- **We do not run formal background screening as a standard step**, and we are not going to describe an\r\n informal hiring process as vetting. Where your own contract or regulatory position requires screening for\r\n named individuals working on your account, tell us at legal@ever.co and we will arrange it.\r\n- **We do not run a certified annual security awareness training programme.** Security expectations are\r\n part of onboarding and part of how we review changes, but that is a practice, not a curriculum, and the\r\n two should not be confused on a page like this.\n\n## Providers and vendor management\r\n\r\n### The first question is whether we need one at all\r\n\r\nWe run our own infrastructure, and that is a security decision as much as a cost one: **a provider we do\r\nnot engage is a provider that cannot be breached on our behalf.** Before we take on a new one we ask\r\nwhether the function can be served by something we already operate. Often it can, and the answer is no\r\nprovider.\r\n\r\n### Before we engage one\r\n\r\n- **What data would it see**, and can we send it less — or none — and still get the benefit?\r\n- **What is its security posture**: its own certifications and reports, its published architecture, its\r\n breach history, how it handles vulnerability reports.\r\n- **Where does it process**, which establishment contracts with us, and which transfer mechanism applies if\r\n processing happens outside the European Economic Area.\r\n- **Will it sign a data processing agreement** meeting Article 28 GDPR, including the Standard Contractual\r\n Clauses where they are needed? If it will not, we do not engage it for anything involving personal data.\r\n- **How do we get out** — can we export what it holds, and can we replace it without rebuilding the\r\n product around its absence?\r\n\r\n### What the contract has to contain\r\n\r\nProcessing only on our documented instructions; confidentiality binding its personnel; security measures at\r\nleast as strict as our own; controls on its own sub-processors; assistance with data subject requests and\r\nwith security, breach notification and impact assessments; notification of a breach without undue delay;\r\ndeletion or return of the data when the engagement ends; and our right to the information needed to\r\ndemonstrate compliance.\r\n\r\n**A provider may not use your data for its own purposes.** Not to improve its own products, not to train\r\nmodels, not to build advertising profiles.\r\n\r\n### While it is engaged\r\n\r\nWe review the providers we use periodically, watch for security incidents and for changes in the legal\r\nbasis for transfers to them, and remove one when the feature it served is gone. A provider that is no\r\nlonger used does not stay on the list as decoration, and a provider that has been quietly replaced is a\r\nchange we announce.\r\n\r\n### How this appears to you\r\n\r\nEvery provider that processes personal data for the hosted Service is published on our\r\n[sub-processor page](https://gauzy.co/subprocessors), in three tiers: those always engaged, those engaged\r\nonly when you enable a specific feature, and those that appear only in a deployment you run yourself. That\r\npage also gives the advance notice period before we add or replace one, and the procedure for objecting.\n\n## Responsible disclosure\r\n\r\nIf you have found a security vulnerability in the Service, we want to hear about it, and we would rather\r\nhear about it from you than from someone using it.\r\n\r\n### How to report\r\n\r\nSend it to **abuse@ever.co**. If personal data is exposed, copy privacy@ever.co.\r\n\r\n**We do not operate a separate security mailbox, and we are not going to publish one we do not monitor.**\r\nThat address reaches the people who operate the Service.\r\n\r\nA useful report contains: what the issue is, the exact URL, endpoint or component, the steps to reproduce\r\nit, what an attacker could actually do with it, and anything you need from us to prove it. Screenshots or a\r\nshort recording help. Write in English.\r\n\r\n### What we do with it\r\n\r\n- **We acknowledge your report within three business days.**\r\n- We assess and reproduce it, and tell you what we found — including if we conclude it is not a\r\n vulnerability, and why.\r\n- We keep you updated while we fix it, and tell you when the fix is live.\r\n- We are happy to **credit you publicly** when the fix has shipped, or to leave you out of it. Your choice.\r\n\r\nSerious issues jump the queue. We fix a critical vulnerability as quickly as we can build and verify a fix,\r\nand we will tell you what the timeline looks like rather than leaving you guessing.\r\n\r\n### Our commitment to you\r\n\r\n**If you follow the rules below, we will not pursue legal action against you, we will not report you to\r\nanyone, and we will not ask your employer to.** We will treat your research as authorised access for the\r\npurposes of computer misuse legislation and our own [Terms of Service](https://gauzy.co/tos), and we will\r\ntreat your report as a contribution rather than a threat.\r\n\r\nThis protection covers you personally. It cannot cover you against a third party — if your testing reaches\r\none of our providers' systems, that is between you and them.\r\n\r\n### The rules\r\n\r\n- **Do not access, modify, delete or download anyone else's data.** Use your own test account. If you can\r\n reach another customer's data, that is the finding — stop there and tell us. You do not need to collect\r\n it to prove it.\r\n- **Stop as soon as you have demonstrated the issue.** Do not pivot, escalate further, or establish\r\n persistence.\r\n- **No denial of service**, volumetric testing, load testing or resource exhaustion.\r\n- **No social engineering** of our people, our customers or our providers, and no physical attempts against\r\n our premises or equipment.\r\n- **No spam, no mass account creation, no brute forcing** of live accounts.\r\n- **Give us a reasonable time to fix it before you publish** — 90 days from your report, or the day the fix\r\n ships, whichever comes first. If a fix is taking longer, talk to us; we will not use the clock to bury a\r\n finding.\r\n- **Obey the law**, and do not do anything that would harm the Service or the people who use it.\r\n\r\nReports produced by running an automated scanner against the live Service are not in scope, and running one\r\nis itself a breach of the testing section above.\r\n\r\n### Usually out of scope\r\n\r\nMissing security headers or best-practice settings with no demonstrated impact; self-inflicted\r\ncross-site scripting; missing rate limits without a working attack; outdated version banners without a\r\nworking exploit; results from a vulnerability scanner submitted without validation; social engineering;\r\nand issues in a third-party service rather than ours — report those to the provider that runs it.\r\n\r\nSend it anyway if you are unsure. We would rather triage a duplicate than miss a real one.\r\n\r\n### There is no paid bounty\r\n\r\n**We do not run a paid bug bounty programme and we do not pay for vulnerability reports.** We would rather\r\nsay that plainly at the start than have you invest a weekend on the assumption that we do. What you get is\r\na fast, straight response from the people who can fix it, and public credit if you want it.\r\n\r\n### Security issues in our open-source code\r\n\r\nWhere our software is published as open source, report a security issue in the code to the same address, in\r\nprivate. **Please do not open a public issue or a pull request that describes the vulnerability** — that\r\npublishes a working exploit to everyone running the code, including people who cannot patch quickly.\n\n## Shared responsibility — who secures what\r\n\r\nSecurity of a hosted service is split between the provider and the customer, and **most real incidents\r\nhappen on the customer's side of that line**: a shared password, a departed employee who kept access, an\r\nintegration token that outlived its purpose, a laptop left on a train. That is not a criticism. It is the\r\nreason this section exists and is specific.\r\n\r\n### What we are responsible for\r\n\r\n- The physical hardware, the facilities and the virtualisation platform.\r\n- The container platform, the operating systems and the runtime the application runs on.\r\n- Patching and updating everything in the list above, and the application itself.\r\n- Encryption in transit and at rest, and the management of our keys.\r\n- Network security, the edge, and keeping the origin off the public internet.\r\n- Availability, redundancy, backups of the Service, and restoring it after a failure.\r\n- Isolation between customers.\r\n- The security of our own code, our build pipeline and our deployment process.\r\n- Our own staff: their access, their devices, their confidentiality obligations.\r\n- The providers we engage, and the contracts that bind them.\r\n- Detecting, responding to and notifying you of incidents in our part.\r\n\r\n### What you are responsible for\r\n\r\n**Who gets in, and who stays in.** Your users, their roles and their permissions. Inviting people, removing\r\nthem the day they leave, reviewing periodically who still has access, and not sharing accounts. Choosing\r\nstrong, unique passwords and turning on multi-factor authentication where the product offers it. If your\r\norganisation uses single sign-on, the security of your identity provider is yours.\r\n\r\n**How you configure the product.** Every switch we give you is a decision you are making. Defaults are\r\nchosen to be safe, not to be right for you, and changing one is a deliberate act with consequences you own.\r\n\r\n**Your monitoring configuration.** For products that can capture workforce activity, **you decide which\r\ncapture features are switched on, how often they run, and who can see the result.** We supply the controls;\r\nwhich of them is lawful and proportionate in your organisation, in your jurisdiction, is your call. The\r\nnext block is about that in more detail, because it is the sharpest edge in this document.\r\n\r\n**Your retention settings.** Where the product lets you set how long data is kept, that period is yours.\r\nKeeping data longer than you need it is a risk you are choosing to hold, and it is one of the first things\r\na supervisory authority asks about.\r\n\r\n**Your integrations and your tokens.** Every third-party service you connect, every API key you issue,\r\nevery webhook you point somewhere and every access token you grant to an application or an automated agent.\r\nApproving an integration is you deciding to send your data somewhere else. Scope those grants narrowly,\r\nreview them, and revoke them when a person leaves or a project ends — a token issued by someone who has\r\nleft your organisation still works until you revoke it.\r\n\r\n**Your devices and your endpoints.** The laptops, phones and desktops your people use to reach the Service,\r\nincluding any desktop or mobile application we publish. Full-disk encryption, screen locks, patching,\r\nmalware protection and what happens when a device is lost are yours. We cannot protect an account from a\r\ncompromised machine that is already signed into it.\r\n\r\n**Your data, and what you put in it.** What you choose to upload, whether it belongs in this product at\r\nall, and whether you have the right to process it. Also your own copies — see the backups section.\r\n\r\n**Telling your own people.** Your privacy notice to your staff, customers and site visitors is yours to\r\nwrite. Ours describes what we do; it is not your notice, and our cookie banner is not your consent\r\nmechanism.\r\n\r\n### Where you use the Service to monitor people\r\n\r\nSome of our products capture information about how an organisation's own personnel work — which can include\r\nscreenshots, application and website usage, activity rates derived from keyboard and mouse events, and,\r\nwhere you switch them on, webcam stills, audio or screen recordings.\r\n\r\nFor all of that data, **you are the controller and we are your processor.** You decide why it is collected\r\nand how; we process it on your instructions. That split has consequences you should read carefully:\r\n\r\n- **The lawful basis is yours to establish.** Consent is generally not a valid basis in an employment\r\n relationship because of the imbalance of power between employer and employee, so a signed form is\r\n unlikely to be the answer you need.\r\n- **The notice to the monitored person is yours to give**, in advance, covering each type of capture that\r\n is switched on.\r\n- **The data protection impact assessment is yours to complete** where one is required, which it commonly\r\n is for systematic monitoring of employees. We will help: ask privacy@ever.co for the information you\r\n need about data categories, capture frequencies, retention, storage locations, sub-processors and\r\n transfer mechanisms, and we will provide it.\r\n- **Consultation with a works council, trade union or employee representative body is yours to run** where\r\n local law requires it — which it does in several European countries, before the system is introduced, not\r\n after.\r\n- **Some capture types are restricted or prohibited in some countries.** Continuous screen capture in\r\n particular is treated very differently across Europe.\r\n\r\n**We make no claim that any monitoring feature is lawful for you to use, anywhere.** A feature existing in\r\nthe product is not advice that you may switch it on, and our providing it is not our assessment of your\r\nlegal position. Where you enable one, you confirm to us that you have done the work above — the Data\r\nProcessing Addendum and the [Terms of Service](https://gauzy.co/tos) put that in binding terms, including\r\nwho answers for it if it turns out not to have been done.\r\n\r\n### Where you run the software yourself\r\n\r\nIf you deploy our open-source software on your own infrastructure, **everything above is yours** — the\r\nservers, the database, the storage, the backups, the encryption, the patching, the access control and the\r\nincident response. You are the controller of the data in it. We have no access to it, we are not your\r\nprocessor for it, and no commitment on this page applies to it.\n\n## How to reach us about security\r\n\r\n- **A vulnerability you have found** — abuse@ever.co. Read the responsible disclosure section first; it\r\n tells you what protection you have and what rules go with it.\r\n- **An incident, a compromised account, or something you can see that you should not be able to see** —\r\n abuse@ever.co, immediately. Copy privacy@ever.co if personal data is involved.\r\n- **A security questionnaire, a due diligence request, a request for the Data Processing Addendum, evidence\r\n of a restore test, or permission to test the Service** — legal@ever.co.\r\n- **A question about personal data, a data subject request, or a breach affecting you** —\r\n privacy@ever.co.\r\n\r\nAll of these addresses are monitored by the people who operate the Service. By post: **Ever Technologies LTD**,\r\nMladost 2, bl. 211, ent. A, Sofia 1799, Bulgaria — registered in Bulgaria under company number 204599535. We\r\ncorrespond in English.\r\n\r\nYou may also complain to a data protection supervisory authority. Ours is the\r\nCommission for Personal Data Protection (Комисия за защита на личните данни), the CPDP, at\r\n[https://www.cpdp.bg/](https://www.cpdp.bg/). You may instead complain\r\nto the authority for the country where you live or work.\r\n\r\nThis document is version 1.0.3 of the Security Overview for\r\n[gauzy.co](https://gauzy.co), in force from 2026-10-24. It describes the measures in\r\nplace at that date. Earlier versions, with the dates they applied, are at\r\n[https://gauzy.co/security](https://gauzy.co/security)."
380
380
  }
@@ -8,7 +8,7 @@
8
8
  "locale": "en",
9
9
  "version": "1.0.2",
10
10
  "effectiveDate": "2026-08-02",
11
- "sha256": "94da1b30029d018d9d06c60339d02a891c8a1ca8733bc8b573b0614c3eedc33b",
11
+ "sha256": "f51c105ce3b44d6909c1400582082834c346a9c6ec94f3292f37408d39ee4fa5",
12
12
  "publishable": true,
13
13
  "sections": [
14
14
  "core/security/010-overview",
@@ -417,5 +417,5 @@
417
417
  }
418
418
  ],
419
419
  "path": "content/security/githands/en.html",
420
- "markdown": "## Security overview\n\nThis page describes how we protect GitHands Cloud and the data you keep in it. It is written to be\nchecked rather than admired: every measure below is something we actually run, and where we do not do\nsomething that a page like this usually claims, **we say so instead of leaving a gap you have to guess at**.\n\nThe short version:\n\n- We run the Service on **hardware we own and operate ourselves**, not as a tenant in someone else's cloud.\n The database, the object storage and the secrets store are ours. That is why the list of providers with\n access to your data is short.\n- Traffic is encrypted in transit, storage is encrypted at rest, and off-site backup copies are encrypted\n **before they leave our network**, so the provider holding them holds ciphertext.\n- Production changes go through version control and review. Hand edits on live systems are the exception,\n not the routine.\n- **We hold no security certification, and we do not claim one.** The next section says exactly what we do\n instead.\n\n### What this page covers\n\nIt covers the hosted Service we operate at [githands.com](https://githands.com) — the application, its\ninterfaces, and the infrastructure underneath them.\n\nIt does **not** cover two things that are easy to confuse with it:\n\n- **A copy of the software you run yourself.** Where our software is published as open source and you deploy\n it on your own infrastructure, you are the operator. You choose its hosting, its storage and every other\n component, we have no access to it, and nothing on this page describes it.\n- **Your own systems and your own configuration.** Which of your people have access, how you configure the\n product, what you connect it to and how you secure the devices your team uses are yours. The shared\n responsibility section near the end of this page sets out that split in detail, because most real\n incidents happen on one side of that line rather than in the middle of it.\n\n### What this page is, legally\n\nThis is a factual description of the measures in place at the version and date shown at the end. It is\n**not a warranty, and it does not create a service level commitment.** Our contractual security obligations\nare in our Data Processing Addendum and in the [Terms of Service](https://githands.com/tos), and those are the\ndocuments that bind us.\n\nWe change these measures as the systems and the threats change. **We will not make a change that materially\nreduces the protection described here**, and when the description changes we republish this page with a new\nversion and date.\n\n### One thing we will not tell you\n\n**No system is perfectly secure, and a vendor who tells you otherwise is selling something.** Everything\nbelow reduces risk; none of it eliminates it. What we can commit to is that we take these measures\nseriously, that we test the ones that can be tested, and that when something goes wrong we tell you rather\nthan manage the news.\n\n## Certifications — what we hold, and what we do not\n\n**We do not hold a SOC 2 report of any type. We are not certified to ISO/IEC 27001. We hold no other\nsecurity certification or attestation, and we do not claim one.**\n\nWe are stating that first, in plain words, because a security page that leaves the question open is doing\nso on purpose. If a proposal, a sales conversation, a questionnaire response, a directory listing or a\nthird-party summary has told you that we are SOC 2 or ISO 27001 certified, **it is wrong** — please send it\nto legal@ever.co so we can get it corrected at the source.\n\n### What we do instead\n\nWe use two published control sets as a working checklist rather than as a badge:\n\n- **The control areas of ISO/IEC 27001 Annex A** — access control, cryptography, operations security,\n supplier relationships, incident management, continuity — as the structure for reviewing our own\n practices. This page is organised broadly along those lines.\n- **Article 32 GDPR**, which is not optional for us and is the standard our customers can actually hold us\n to: technical and organisational measures appropriate to the risk, kept under review as the risk changes.\n\n**No external auditor has examined either.** These are self-assessments, and you should read them as such.\nAn unaudited framework alignment and a certification are different things, and we will not blur the two by\nwriting \"aligned with\" in a way that reads like \"certified to\".\n\n### What you can have instead of a certificate\n\nA certificate is a proxy for evidence. Where we cannot give you the proxy, we will give you the evidence:\n\n- **This page**, which is specific enough to be checked against what we actually do.\n- **The security measures schedule in our Data Processing Addendum**, which is contractual rather than\n descriptive — if a measure matters to you, that is where to pin it down.\n- **A completed security questionnaire.** Send yours to legal@ever.co. We answer them, we answer them\n honestly, and where the honest answer is \"no\" you will get \"no\" rather than a paragraph that avoids the\n question.\n- **Evidence on request** for specific points — a description of the architecture, the outcome of our most\n recent restore test, the current sub-processor list, the transfer mechanism for a named provider.\n- **A right to audit**, on the terms set out in the Data Processing Addendum.\n\n### If that changes\n\nIf we obtain a certification or an independent report, this page will say which one, who issued it, what\nits scope covers and what it excludes — because the scope is the part that matters and the part most often\nleft out. Until that day, the sentence at the top of this section is the whole answer.\n\n## Encryption\n\n### In transit\n\nAll traffic between you and the Service travels over TLS. HTTPS is enforced and plain HTTP requests are\nredirected rather than served, so a request that arrives unencrypted never reaches the application with its\ncontents intact. We keep the accepted protocol versions and cipher suites current and retire weak ones as\nthe ecosystem does.\n\nTraffic between our own sites, and between our infrastructure and the edge network in front of it, runs\nover encrypted connections. Traffic to the providers we engage runs over TLS to their published endpoints.\n\n### At rest\n\n- **Storage holding customer content, databases and backups is encrypted at rest.** That protects against\n the loss, theft or improper disposal of physical media — it does not protect against a compromise of a\n running system that already has the data decrypted, and we would rather you knew which threat it answers.\n- **Off-site backup copies are encrypted before they leave our network**, with keys we hold and manage\n ourselves. The provider storing those copies holds ciphertext and nothing else: it cannot read them, and\n neither can anyone who obtains them from it.\n\n### At the application layer\n\nSome values are too sensitive to rely on storage-level encryption alone, so they are encrypted again inside\nthe application before they are written:\n\n- **Credentials, access tokens and API keys** — including tokens you grant us for a third-party integration.\n- **Secrets used to operate the Service** — held in a secrets manager we run ourselves, injected into\n workloads at runtime, and **never committed to source control**.\n\n**Passwords are never stored in a form we can read.** They are put through a modern one-way hashing\nfunction with a per-password salt. We cannot recover your password, tell you what it is, or send it back to\nyou — a reset is the only route, and any service that offers you the alternative is storing it wrongly.\n\nWhere a specific product stores a particular credential differently from the description above, its annex\nto this page says so and says why. **We would rather publish an awkward exception than a tidy sentence that\nis not true everywhere.**\n\n### Keys\n\nEncryption keys are held in our self-hosted secrets manager, separately from the data they protect, and are\naccessible only to the systems and people that need them. Keys are rotated when a person with access leaves,\nwhen we have reason to think a key has been exposed, and on a periodic schedule. **The keys for off-site\nbackups are held independently of the infrastructure they protect**, so that losing a site does not also\nlose the ability to decrypt its backups.\n\n## Access control\n\n### Your access to your own data\n\nAccess inside your workspace is yours to configure. The Service gives you named user accounts, roles and\npermissions, workspace administrators who can invite and remove people, and session controls. Where a\nproduct supports single sign-on, multi-factor authentication or IP restrictions, its annex says so and its\ndocumentation says how to switch them on.\n\n**Turning those controls on, and reviewing who still needs access, is your job — see the shared\nresponsibility section.** The most common way an account is compromised is not a flaw in our systems; it is\na shared password, a former employee who was never removed, or an integration token that outlived its\npurpose.\n\n### Our access to production systems\n\n- **Named individual accounts, never shared logins.** Every action in production is attributable to a\n person.\n- **Multi-factor authentication** on the systems and administrative accounts that support it.\n- **Least privilege.** Access is scoped to the narrowest role that lets the work happen, and elevated\n access is granted for a purpose rather than held permanently by default.\n- **Changes go through Git.** The configuration of the live environment is held in version control and\n applied automatically, so a change is reviewable, attributable and reversible. Direct hands-on changes to\n a running system are an exception for incidents and hands-on maintenance, not the daily routine — and the\n automation reverts undeclared edits rather than leaving them in place.\n- **Removal on departure or role change.** Access is revoked when someone leaves or no longer needs it, and\n credentials they held are rotated. Access is reviewed periodically as well, because revocation lists rot.\n\n### Access to your content\n\nWe access the content in your workspace only where it is necessary to operate, secure or support the\nService — for example to investigate a fault you have reported, or to respond to a security incident. **We\ndo not browse customer content, and we do not use it to build products, train models or produce insights\nfor anyone else.**\n\nSupport access is granted for the task in hand and logged. Where a product offers a mechanism for you to\ngrant or revoke support access yourself, its annex describes it.\n\n### Being honest about the limits\n\n**We are a small team, and several of us hold administrative access because operating this infrastructure\nrequires it.** A large organisation would separate the person who approves a change from the person who\nmakes it; at our size that separation is partial, and we are not going to describe it as complete. What we\ndo instead is make every change visible: production state comes from version control, actions are\nattributable to named accounts, and administrative events are logged and monitored.\n\n## Infrastructure and hosting\n\n### We run our own infrastructure\n\nThe Service does not run in a rented public cloud account. It runs on **physical servers we own, in\nfacilities we control, inside the European Union**, on a virtualisation and container platform we operate\nourselves. The database, the object storage, the message and cache layers, the secrets store and the\ndeployment system are all components we run — not managed services bought from a third party.\n\nThat is an unusual answer in this market, and it has a direct consequence you can verify: **the list of\nproviders with access to your data is short.** Most of what would otherwise be a sub-processor is simply a\nservice we operate.\n\nIn outline, and deliberately without the detail that would help someone attack it:\n\n- **Hardware and virtualisation we own**, clustered so that losing a machine does not lose a service.\n- **A container orchestration platform** running the applications, with the control plane replicated across\n multiple machines.\n- **PostgreSQL**, run by an operator that handles replication, failover and continuous archiving of the\n write-ahead log.\n- **S3-compatible object storage** we host, for files and media, with data replicated across nodes.\n- **A secrets manager we host**, from which credentials are injected into workloads at runtime.\n- **Declarative configuration held in Git** and applied automatically, so the live environment matches a\n reviewed commit rather than an accumulation of manual changes.\n\nA content delivery and security network sits in front of all of it, and our source code, build pipeline and\ncontainer registry are hosted with a third-party provider. Those are genuine third parties even though the\ninfrastructure behind them is ours, and they are named on our\n[sub-processor page](https://githands.com/subprocessors).\n\n### Redundancy and failure domains\n\nServices that can run on more than one machine run on more than one machine. Databases are replicated,\nobject storage is replicated across nodes, and the systems that hold backups run on **a physically separate\ncluster from the systems they protect** — so a failure that takes out the live environment does not take\nits backups with it.\n\n**This is not the same as multi-region redundancy, and we will not describe it as such.** We can lose\nmachines without losing data or availability. A whole-site failure means recovering from off-site copies,\nwhich is a matter of hours rather than seconds. The continuity section is specific about what that means.\n\n### Separation between customers\n\nIsolation between customers is **logical, not physical**, unless you have separately contracted for a\ndedicated deployment. Depending on the product that means a separate database per customer, a separate\nnamespace and deployment per customer site, or tenant scoping enforced in the application and the data\nmodel. **Your product's annex states which model applies to you** — it is a question worth asking of any\nvendor, and worth a straight answer.\n\n### Separation between environments\n\nDevelopment, staging and production are separate environments with separate credentials and separate data\nstores. Production credentials are not available in development or staging, and **we do not copy production\ndata into a development environment**. Test data is synthetic or anonymised.\n\n### Third-party managed services, where they are used\n\nA small number of products use a third-party managed database, managed storage or hosted platform for a\nspecific function rather than our own. Where that is the case it is named in that product's annex and it\nappears in the always-engaged tier of the sub-processor list. **We do not describe the estate as fully\nself-hosted where a product's data actually sits somewhere else.**\n\n## Network security\n\n### At the edge\n\nEvery request to the Service passes through a content delivery and security network before it reaches us.\nThat layer terminates TLS, absorbs volumetric denial-of-service traffic, applies a web application firewall\nand filters automated abuse. It also means the address of our origin infrastructure is not the address your\nbrowser connects to.\n\n### The origin is not exposed\n\n**Our application servers do not publish inbound ports to the public internet.** Traffic reaches them\nthrough outbound-initiated tunnels established from inside our network to the edge, so there is no\npublic listener to scan, and no direct path from the internet to an origin host.\n\n### Inside the network\n\n- Databases, object storage, the secrets manager and internal services sit on **segmented private networks\n and are not reachable from the internet**. They accept connections only from the workloads that need them.\n- **Administrative interfaces are not published publicly.** Reaching them requires being on the internal\n network, through a controlled route with authenticated access.\n- Outbound connectivity from workloads is limited to what the workload actually needs, so a compromised\n component has fewer places to send data and fewer places to fetch a second stage from.\n\n### Abuse and automation\n\nSign-up, sign-in, password reset and contact endpoints are protected against automated abuse by rate\nlimiting and by a bot-detection challenge. **That challenge is strictly necessary to keep the Service\nworking and available** — it protects accounts from credential-stuffing and forms from bulk abuse — which\nis why it runs without a consent prompt, as explained in our [Cookie Policy](https://githands.com/cookies). It\nis not used for analytics, profiling or advertising.\n\n### What we monitor for\n\nUnusual authentication patterns, unexpected administrative activity, sudden changes in traffic shape and\nfailures of the components above raise alerts that a person sees. The logging and monitoring section\ndescribes what is recorded and for how long.\n\n## Application security\n\n### How a change reaches production\n\nEvery change to the Service starts as a change to source code in version control and follows the same\nroute:\n\n1. **A pull request**, reviewed by someone other than its author on the branches that matter.\n2. **Automated checks** — build, tests, static analysis and dependency scanning — which must pass.\n3. **Promotion through environments.** Changes move development to staging to production, in that order.\n Production is deployed from the release branch and from nothing else.\n4. **A container image built by the pipeline** from a reviewed commit, identified by its digest, deployed\n by the same automation that holds the environment's configuration in Git.\n\n**Nobody edits production by hand as a way of shipping a change.** The deployment system reconciles the\nlive environment against the version-controlled definition, so an undeclared edit is reverted rather than\nquietly kept.\n\n### Automated checks on every change\n\n- **Dependency scanning**, with alerts for known vulnerabilities in the packages we use, and updates\n applied on a defined schedule with a faster path for serious issues.\n- **Static analysis** of our source code, run on each change, flagging the classes of defect it can find\n before the code ships.\n- **Secret scanning** across our repositories, with pushes blocked where a credential is detected. Where a\n credential has been exposed, we treat rotating it at the provider as the fix — deleting it from the code\n is not the fix.\n- **Automated tests**, which have to pass before a change can be merged or released.\n\nDependencies are installed through an internal package cache we run, which gives us a consistent, auditable\nview of what actually went into a build.\n\n### In the application itself\n\n- Database access goes through parameterised queries and a mapping layer rather than assembled strings.\n- Output is encoded for its context, and state-changing requests are protected against cross-site request\n forgery.\n- Authentication, session handling and password storage use established libraries rather than anything we\n invented; passwords are hashed as described in the encryption section.\n- Authorisation is enforced on the server for every request. A hidden control in the interface is not an\n access control.\n- Input is validated at the boundary, and uploads are constrained by type and size.\n- Browser-facing responses carry security headers, including a content security policy on the surfaces\n where one can be applied without breaking the product.\n\n### The honest limit\n\n**Automated tooling finds categories of defect, not all defects.** Review catches more, and neither catches\neverything. That is why the sections on logging, incident response and responsible disclosure exist: they\nare what we rely on for the bugs that get through, and something always gets through.\n\n## Testing\n\n### What we test, and how often\n\n- **Automated test suites** run on every change, in the pipeline, before anything can be merged or\n released.\n- **Static analysis and dependency scanning** run on every change as well, as described in the previous\n section.\n- **Restore tests.** We restore backups into a scratch environment and check the result against the source\n rather than assuming a backup that completed is a backup that works. This is the test most often skipped\n in the industry and the one most often needed.\n- **Manual review** of changes that touch authentication, authorisation, payment flows, data export or the\n handling of credentials — the areas where an automated check is least likely to notice that the logic is\n wrong.\n- **Configuration review** of the version-controlled definition of the live environment, which is where a\n surprising share of real exposure lives.\n\n### Penetration testing — what we do not have\n\n**We have not commissioned an independent penetration test of the hosted Service, and we do not claim one.\nThere is no test report for us to send you, and any summary suggesting otherwise is not ours.**\n\nWe would rather write that sentence than the usual formulation about \"regular security assessments\" that\nturns out on inspection to mean an automated scanner. When we do commission a test, this page will say who\nperformed it, when, what was in scope and what was excluded.\n\n### Testing the Service yourself\n\n**You may not run security testing against the Service without our written agreement** — automated\nscanning, load testing, fuzzing and exploitation attempts included. From our side an unannounced scan is\nindistinguishable from an attack, and it will be handled as one.\n\nIf you want to test, we will usually say yes. Write to legal@ever.co and we will agree the scope, the\nwindow, the accounts you use and how findings are reported. The conditions are the obvious ones: your own\ntest data only, no other customer's data touched, nothing that degrades the Service for anyone else, and\nfindings to us before they go anywhere else.\n\nIf you have already found something without meaning to, do not test further — go straight to the\nresponsible disclosure section, which is written for exactly that situation and offers you protection for\nit.\n\n## Logging and monitoring\n\n### What we record\n\n- **Authentication events** — sign-in, sign-out, failed attempts, password and multi-factor changes.\n- **Administrative and security-relevant events** — permission changes, invitations and removals, token\n creation and revocation, changes to workspace settings, access to production systems.\n- **Application programming interface access** — the request, its source, its outcome and its timing.\n- **Infrastructure and platform events** — deployments, configuration changes, node and service state.\n- **Application errors and diagnostics** — enough to work out what broke.\n\n### What we try not to record\n\nWe do not deliberately log passwords, session tokens, API keys or payment card numbers, and where such a\nvalue would otherwise appear in a log or an error report it is redacted. **We will not claim this is\nperfect** — a value can end up in a stack trace nobody predicted. When we find one, we treat it as an\nexposure: the credential is rotated and the log entry removed.\n\nWhere a product sends error reports to a third-party error-tracking service, that service is named in the\nproduct's annex and on the [sub-processor page](https://githands.com/subprocessors), because an error report can\ncarry an identifier for the person who hit the error.\n\n### How long we keep it\n\n| Log type | Retention |\n|---|---|\n| Security, access and audit logs | **12 months** |\n| Application and error logs, crash reports, diagnostics | **90 days** |\n\nThese are the same periods published in the retention table of our\n[Privacy Policy](https://githands.com/privacy), and if the two ever disagree the Privacy Policy governs.\n\nAccess to logs is restricted to the people whose job needs it, and log access is itself an administrative\nevent that gets recorded.\n\n### Monitoring and alerting\n\nMetrics from the platform, the databases, the storage layer and the applications feed a monitoring system\nthat raises alerts when something is wrong — a service failing, a certificate close to expiry, a backup\nthat did not complete, a disk filling, an unusual pattern of authentication failures. **Alerts go to people,\nnot to a dashboard nobody opens.**\n\nWe also run an **independent external check** that watches the Service and the monitoring system itself\nfrom outside our network, and raises an alarm if either goes quiet. Monitoring that cannot survive the\nfailure it is meant to detect is decoration, and a failed backup that alerts nobody is the standard way a\nrecovery turns out to be impossible.\n\n## Incident response\n\n### How we handle an incident\n\n1. **Detect and record.** An alert, a customer report or a disclosure from a researcher opens an incident,\n with one named person responsible for running it.\n2. **Assess and classify.** What happened, what is affected, whether personal data is involved, whose data,\n and how bad it could get. Severity drives everything that follows.\n3. **Contain.** Cut the access, revoke the credential, isolate the component, block the route — stopping it\n comes before understanding it.\n4. **Eradicate and recover.** Remove the cause, restore service, restore data from backup where that is\n what is needed, and verify the result rather than assuming it.\n5. **Notify.** Under the timescales below.\n6. **Review afterwards.** A written review of what happened, why, what we changed, and what would have\n caught it sooner.\n\n### Telling you about it\n\n**Where we are the controller** — for the data described in our [Privacy Policy](https://githands.com/privacy) as\nours — and there is a personal data breach that is likely to result in a risk to people:\n\n- We notify the CPDP **within 72 hours of becoming aware of it**, as\n Article 33 GDPR requires, and where we cannot give complete information in that window we give what we\n have and follow up.\n- We notify the affected people **without undue delay** where the breach is likely to result in a high risk\n to them.\n\n**Where we are your processor** — everything inside your workspace, including data your organisation\ncollects about its own personnel — you are the controller and the notification duty to the authority is\nyours, so ours is to you:\n\n- We notify you **without undue delay after we become aware, and in any event within 72 hours.**\n- We give you what Article 33(3) requires you to have: the nature of the breach, the categories and\n approximate number of people and records involved, the likely consequences, the measures we have taken or\n propose, and a contact point who can answer follow-up questions.\n- We assist you with your own notifications and with any questions from your supervisory authority. Our\n Data Processing Addendum sets that out contractually.\n\nWe notify you at the administrative contact addresses on your account. **Keeping those current is worth\nfive minutes of your time**, because a notification that reaches a mailbox nobody reads has been made and\nnot received.\n\n### What we will not do\n\n- **We will not sit on a breach** to work out how it looks, or wait until we have a complete picture before\n telling you something is wrong.\n- **We will not require you to sign anything** — a non-disclosure agreement, a waiver, an amended contract —\n as a condition of being told.\n- **We will not describe an incident as something else.** If data was accessed, the notice says data was\n accessed.\n\n### Reporting an incident to us\n\nIf you think an account has been compromised, credentials have leaked, or you have seen something in the\nService you should not be able to see, tell us immediately at abuse@ever.co. If personal data is\ninvolved, copy privacy@ever.co. Both addresses are monitored and reach the people who operate the\nService.\n\nIf you have found a vulnerability rather than an incident in progress, the responsible disclosure section\nbelow is the route, and it offers you protection that this section does not.\n\n## Continuity and backups\n\n### Three tiers, and why there are three\n\nBackups fail in three different ways — the copy is too old, the copy is on the machine that died, or the\ncopy is in the building that burned. So there are three tiers, each answering a different failure:\n\n- **Tier 1 — a fast local copy.** Recent database backups are staged on storage separate from the live\n database, on a different machine. This is the tier that answers \"somebody dropped the wrong table twenty\n minutes ago\" and it restores quickly because the data never left the site.\n- **Tier 2 — a durable on-site copy.** Everything is copied into a **private, backup-only object store that\n runs on a physically separate cluster from the systems it protects**, with its own credentials, reachable\n only from inside our network and never from the internet. Losing the cluster that runs the Service does\n not lose the cluster that holds its backups.\n- **Tier 3 — an off-site copy with a third party.** A further copy goes to object storage operated by an\n external provider, in a different place entirely, **encrypted before it leaves our network**. The\n provider stores ciphertext and holds no key. This is the tier that answers the loss of the whole site.\n\nDeliberately, **the credentials that run the Service are not the credentials that can delete its backups.**\n\n### What is backed up\n\n- **Databases** — continuous archiving of the write-ahead log for point-in-time recovery, plus periodic\n full backups and separate logical dumps. Two mechanisms, because they fail differently.\n- **Object storage** — the files and media held for you, mirrored to the backup tiers.\n- **Cluster and platform state**, including the configuration needed to rebuild the control plane.\n- **The secrets store**, snapshotted on a short cycle.\n- **The version-controlled configuration** that defines the environment, which is what makes a rebuild\n possible rather than merely theoretical.\n\n**Not backed up, on purpose:** caches and queues, which are disposable by design and reconstructed on\nrestart, and build artefacts that the pipeline can regenerate. Backing those up would add cost and\nrestore time for nothing.\n\n### Recovery objectives\n\nThese are the objectives we design and test against. **They are internal targets, not contractual\ncommitments** — no service level applies unless we have separately agreed one with you in writing, as the\n[Terms of Service](https://githands.com/tos) says.\n\n| Scenario | Recovery point — how much data is at risk | Recovery time — how long it takes |\n|---|---|---|\n| Loss of a machine or a service | None. Replicated components carry on | Minutes, usually without anyone noticing |\n| Database corruption or an unwanted change | Minutes — continuous archiving supports point-in-time recovery | Hours, depending on the size of the database |\n| Loss of the storage holding files | Last completed replication cycle | Hours |\n| **Loss of the whole site** | Last completed off-site cycle | **Hours to days.** This is the honest number |\n\n**We are candid about the last row.** We are resilient to losing machines; we are not a multi-region\nservice that fails over silently, and we will not describe ourselves as one.\n\n### We test restores\n\nA backup nobody has restored is a guess. We restore into a scratch environment and verify the result\nagainst the source — record counts, integrity checks, and whether the restored system actually starts —\nrather than trusting a job that reported success. Restore drills are documented, and the outcome of the\nmost recent one is available to customers on request at legal@ever.co.\n\nMonitoring alerts on a backup that is late, failing or stale, because **a backup system that fails silently\nis functionally the same as no backup system**, and is discovered at the worst possible moment.\n\n### How long backups are kept, and what deletion means\n\nEach backup ages out on its own rolling cycle, within the period given in the retention table of our\n[Privacy Policy](https://githands.com/privacy). Backups are not individually edited to remove one record — that\nwould defeat what a backup is for — so a deleted record persists in backup copies until that cycle\ncompletes. During that window the data is not used for anything, and **we do not restore a backup in order\nto bring back data somebody asked us to delete.** A backup is restored to recover a system, not to\nresurrect a record.\n\n### Your own copies\n\nWe maintain backups for our operational continuity and for the recovery obligations we owe you. **They are\nnot a substitute for your own copies**, and the Service provides export so that you can keep some. If your\nbusiness could not tolerate the loss of the data you keep with us, export it on a schedule that reflects\nthat.\n\n## Retention, deletion and export\n\n### Getting your data out\n\nYou can export your data while your account is open — through the product's own export function where it\nhas one, through the application programming interface, or by asking us. Exports are in structured,\nmachine-readable formats — commonly CSV and JSON, with files delivered in their original format. Your\nproduct's annex says which export routes exist for it.\n\nIf you need an export we do not offer as a self-service function, write to legal@ever.co and we will\nproduce it. **We do not charge for a reasonable export, and we do not make leaving harder than joining.**\n\n### After your agreement ends\n\nYour data remains available for export for **30 days** after the agreement ends. After that window we\ndelete it. The same period, and the reasoning behind it, is in the retention table of our\n[Privacy Policy](https://githands.com/privacy).\n\nIf you need longer, ask before the window closes rather than after it.\n\n### Deletion on request\n\nWhere you delete something inside the Service, or ask us to delete your workspace:\n\n- **Live systems are updated straight away.** The data is gone from the running service.\n- **Backup copies are not individually edited.** A deleted record persists in backups until each copy ages\n out on its rolling cycle, within the period stated in the Privacy Policy. During that window the data is\n not used for anything, and we do not restore a backup to bring it back.\n- **Copies held by a provider we engage** are deleted on our instruction, under the Article 28 terms in our\n contract with them.\n\nWhere we can keep something useful without keeping anyone identifiable — aggregate usage counts, for\nexample — we anonymise rather than delete, and once anonymised it is no longer personal data.\n\n### Retention you control\n\nFor data inside your workspace we are your processor, and **you set the retention period, not us**. Where a\nproduct lets you configure how long activity data, captured media, logs or archived records are kept, that\nsetting is yours and its default is documented in the annex. Choosing a period that is lawful and\nproportionate in your jurisdiction is part of your controller responsibility — see the shared\nresponsibility section.\n\nOur own retention periods, for the data where we are the controller, are published in full in the\n[Privacy Policy](https://githands.com/privacy) as a table with actual numbers rather than the phrase \"as long as\nnecessary\".\n\n### Legal hold\n\nWhere we are required to preserve data for a legal claim, an investigation or a regulatory request, we keep\nit until that ends, including any appeal period, and we do not delete it in the meantime. That obligation\noverrides the periods above. It applies to us; it will sometimes apply to you as well.\n\n### Disposing of hardware\n\nBecause we own our infrastructure, we also retire our own drives. Storage is encrypted at rest, and a drive\nis wiped, cryptographically erased or physically destroyed before it leaves our control. **A retired drive\nis never sold, donated or returned intact.**\n\n## Personnel\n\nWe are a small, long-standing team, and the people who build the Service are the people who operate it.\nThat has a security upside — nobody is administering a system they do not understand — and a limit we set\nout in the access control section rather than hide.\n\n**Everyone with access to production systems or customer data:**\n\n- is bound by **written confidentiality obligations that continue after their engagement ends**, whether\n they are an employee or a contractor;\n- works on a **need-to-know basis**, with access scoped to their role and granted for a purpose rather than\n by default;\n- uses **named individual accounts with multi-factor authentication**, never a shared login;\n- is briefed on how we expect data to be handled as part of joining, and again when something material\n changes;\n- **loses that access when they leave or when their role changes**, with the credentials they held rotated\n as part of the same step.\n\nContractors and anyone working on our behalf are engaged on terms at least as strict as those above. We do\nnot give a provider or a contractor access to customer data because it is convenient.\n\n### Devices\n\nMachines used to access production systems or customer data run full-disk encryption, lock automatically,\nare kept patched, and are protected by multi-factor authentication on the accounts they hold. A lost or\nstolen device is treated as an incident: access is revoked and credentials are rotated before anything else\nis decided.\n\n### Two things we do not claim\n\n- **We do not run formal background screening as a standard step**, and we are not going to describe an\n informal hiring process as vetting. Where your own contract or regulatory position requires screening for\n named individuals working on your account, tell us at legal@ever.co and we will arrange it.\n- **We do not run a certified annual security awareness training programme.** Security expectations are\n part of onboarding and part of how we review changes, but that is a practice, not a curriculum, and the\n two should not be confused on a page like this.\n\n## Providers and vendor management\n\n### The first question is whether we need one at all\n\nWe run our own infrastructure, and that is a security decision as much as a cost one: **a provider we do\nnot engage is a provider that cannot be breached on our behalf.** Before we take on a new one we ask\nwhether the function can be served by something we already operate. Often it can, and the answer is no\nprovider.\n\n### Before we engage one\n\n- **What data would it see**, and can we send it less — or none — and still get the benefit?\n- **What is its security posture**: its own certifications and reports, its published architecture, its\n breach history, how it handles vulnerability reports.\n- **Where does it process**, which establishment contracts with us, and which transfer mechanism applies if\n processing happens outside the European Economic Area.\n- **Will it sign a data processing agreement** meeting Article 28 GDPR, including the Standard Contractual\n Clauses where they are needed? If it will not, we do not engage it for anything involving personal data.\n- **How do we get out** — can we export what it holds, and can we replace it without rebuilding the\n product around its absence?\n\n### What the contract has to contain\n\nProcessing only on our documented instructions; confidentiality binding its personnel; security measures at\nleast as strict as our own; controls on its own sub-processors; assistance with data subject requests and\nwith security, breach notification and impact assessments; notification of a breach without undue delay;\ndeletion or return of the data when the engagement ends; and our right to the information needed to\ndemonstrate compliance.\n\n**A provider may not use your data for its own purposes.** Not to improve its own products, not to train\nmodels, not to build advertising profiles.\n\n### While it is engaged\n\nWe review the providers we use periodically, watch for security incidents and for changes in the legal\nbasis for transfers to them, and remove one when the feature it served is gone. A provider that is no\nlonger used does not stay on the list as decoration, and a provider that has been quietly replaced is a\nchange we announce.\n\n### How this appears to you\n\nEvery provider that processes personal data for the hosted Service is published on our\n[sub-processor page](https://githands.com/subprocessors), in three tiers: those always engaged, those engaged\nonly when you enable a specific feature, and those that appear only in a deployment you run yourself. That\npage also gives the advance notice period before we add or replace one, and the procedure for objecting.\n\n## Annex: GitHands\n\nEverything above describes how we protect a customer's data. **Most of the personal data in this Service\nbelongs to people who are not customers** — developers enumerated from public platforms, who have no account,\nno password to choose, no multi-factor setting to switch on and no way to check on us. They cannot take a\nsingle one of the precautions the shared-responsibility section asks of you.\n\n**That moves the whole burden onto our side of the line**, and this annex is about the parts of it that are\nspecific to this product.\n\n### The source-platform credential you connect\n\nThe Service reads GitHub, GitLab and similar platforms under a credential **you** connect, in your own name.\nIt is the most sensitive thing you hand us, so here is exactly what happens to it.\n\n**What it is used for.** Reading public data through the platform's own official public API: profiles,\nstargazer and forker lists, contributor and participant lists, organisation followers, public commit\nmetadata, and the dependency manifests in public repositories. **It is not used to write anything, anywhere**\n— no commits, no issues, no comments, no repository changes — and it is not used to read anything that is not\npublic. If your grant permits more than that, we still do less.\n\n**A grant can be wider than the use.** Platforms issue access in coarse blocks, so the token you create may\ncarry more access than the Service needs or exercises. **Review the grant at the platform, give it the\nnarrowest form the platform offers, and revoke it there whenever you want** — revocation takes effect on the\nplatform's side immediately, and the Service simply stops being able to read.\n\n**How it is stored, stated exactly.** Tokens are held in our database, on our own infrastructure, and we are\nin the middle of moving them to application-layer encryption — a value encrypted inside the application\nbefore it is written, as described in the encryption section above. **That migration is not finished. Until\nit is, the token is also still written to the older column that carries no application-layer encryption**, so\nin that path it is protected by storage-level encryption and database access control rather than by\napplication-layer encryption. We are completing the migration and removing the older column. **We would\nrather publish that sentence than a tidier one that is not yet true**, and if you would prefer to wait for\nthe migration before connecting a token, that is a reasonable thing to want — ask legal@ever.co where it\nstands.\n\n**We do not pool credentials.** Your reads run under your credential and nobody else's, and we do not use a\ncustomer's token to serve another customer's work.\n\n### Keys, agents and machine access\n\n- **An API key can do what your account can do.** It is a bearer credential: whoever holds it is you, as far\n as the Service is concerned. Treat it like a password, keep it out of source control, rotate it, and tell\n us at abuse@ever.co the moment one is exposed so we can revoke it.\n- **The command-line client, the machine interface and connected agents all act with your credentials**, and\n what they do is attributed to you.\n- **Outbound webhooks send data to an address you configure.** Whatever is at that address receives it. Point\n them at endpoints you control, over HTTPS, and verify what arrives.\n- **A discovery document describing what the Service can do is served without authentication.** It describes\n capabilities, not people. Reaching any data still requires a credential — **and if you find a route that\n returns information about anyone without one, that is a defect we want reported**, through the responsible\n disclosure route below.\n\n### Where non-users' data actually goes\n\nThe sub-processor list at [https://githands.com/subprocessors](https://githands.com/subprocessors) names every\nprovider. Four of them are worth calling out on a security page, because they hold data about people who\nnever chose us:\n\n- **The AI routing provider** receives the prompt for every AI request — names, usernames, email addresses,\n biographies, repository lists and the drafted message — and forwards it to the model provider it routes to.\n- **The AI observability service** stores the full prompt and the full completion for each call as a readable\n trace, so that same personal data sits there in plain text.\n- **Our error monitoring runs with server-side logging enabled**, which means console output can travel with\n a report and can carry fragments of whatever request failed. That is a real exposure route and not a\n theoretical one, which is why we say it rather than describing \"error reports\" and stopping.\n- **The background-job runtime executes extraction, enrichment, scoring and sending**, so it holds job\n payloads and run logs — and **it holds the credentials it needs to reach our database.** A provider that\n holds a database credential is a provider that can reach application data. It is the highest-value\n dependency in this product, and treating it as merely \"a job runner\" would be misleading.\n\nEach is bound by a written contract, may use the data only on our instructions, and **may not use it for its\nown purposes or to train models on it.**\n\n### Separation between customers\n\nCustomer projects, campaigns and records are separated logically inside shared infrastructure — a shared\ndatabase with tenant-scoped records, not a database per customer. **That is logical separation, not dedicated\ninfrastructure, and we will not describe it as more than it is.** Access is enforced in the application on\nevery request; the general application-security section above describes how we build and review that.\n\n### The public pages are public on purpose\n\nThe community pages name individuals and are reachable without an account. There is nothing to breach there\n— it is published deliberately, and it is disclosed as such in the [Privacy Policy](https://githands.com/privacy)\nannex, together with the route to have an entry taken down. We mention it here so nobody reports it as a\nmisconfiguration, and so nobody assumes the rest of the Service is equally open. **It is not: everything\nabout a customer's projects, campaigns and records requires authentication.**\n\n### Suppression is a security property, not just a preference\n\nWhen someone objects, we keep **a one-way hash of their address and a date, and nothing else** — we cannot\nread the address back out of it, and its only job is to make sure the extraction pipeline never re-adds them.\n\nThat record is integrity-critical. **Anything that lets a suppression be bypassed, overwritten, expired or\nevaded is a security defect in our book**, not a product limitation, and we would rather hear about it from\nyou than from the person who started receiving mail again.\n\nThe same goes in the other direction: **the unsubscribe and objection routes are protected against being\niterated**, because a route that lets somebody walk through identifiers could be used to unsubscribe another\ncustomer's recipients wholesale. If you find a way to do that, tell us.\n\n### What we ask of you\n\nBeyond the general shared-responsibility split:\n\n- **Connect the narrowest credential the platform will issue**, and rotate or revoke it when the person who\n created it leaves your organisation. A token issued by someone who has left still works until it is\n revoked.\n- **Use a sending domain you control**, with SPF, DKIM and DMARC configured. A sending path you do not\n control is a security problem before it is a deliverability one.\n- **Treat exported records as personal data about third parties**, because that is what they are. An export\n sitting in a spreadsheet on a laptop is the most common way this kind of data leaks, and once it has left\n the Service we cannot protect it or delete it for you.\n\n### If you run it yourself\n\nEverything in this document is about the Service we operate. In your own installation, the database, the\nstorage, the keys, the patching, the access control and the incident response are yours.\n\n**One default is worth checking before you go live.** The administrator alert address ships pointing at an\naddress on our domain, so until you change it, alerts about your campaign failures — which can carry\noperational detail about your sending — arrive with us. Set it to your own address. The\n[Privacy Policy](https://githands.com/privacy) annex describes the anonymous heartbeat a self-hosted installation\nsends and how to switch it off.\n\n### If you are in the database rather than a customer\n\n**The responsible disclosure route below is open to you.** You do not need an account, you do not need to be\na customer, and you do not need our permission to look at what is publicly reachable. If you find a way to\nget at data about other people through this Service, tell us at abuse@ever.co and we will treat it as\nwhat it is — the most serious kind of report this product can receive.\n\n## Responsible disclosure\n\nIf you have found a security vulnerability in the Service, we want to hear about it, and we would rather\nhear about it from you than from someone using it.\n\n### How to report\n\nSend it to **abuse@ever.co**. If personal data is exposed, copy privacy@ever.co.\n\n**We do not operate a separate security mailbox, and we are not going to publish one we do not monitor.**\nThat address reaches the people who operate the Service.\n\nA useful report contains: what the issue is, the exact URL, endpoint or component, the steps to reproduce\nit, what an attacker could actually do with it, and anything you need from us to prove it. Screenshots or a\nshort recording help. Write in English.\n\n### What we do with it\n\n- **We acknowledge your report within three business days.**\n- We assess and reproduce it, and tell you what we found — including if we conclude it is not a\n vulnerability, and why.\n- We keep you updated while we fix it, and tell you when the fix is live.\n- We are happy to **credit you publicly** when the fix has shipped, or to leave you out of it. Your choice.\n\nSerious issues jump the queue. We fix a critical vulnerability as quickly as we can build and verify a fix,\nand we will tell you what the timeline looks like rather than leaving you guessing.\n\n### Our commitment to you\n\n**If you follow the rules below, we will not pursue legal action against you, we will not report you to\nanyone, and we will not ask your employer to.** We will treat your research as authorised access for the\npurposes of computer misuse legislation and our own [Terms of Service](https://githands.com/tos), and we will\ntreat your report as a contribution rather than a threat.\n\nThis protection covers you personally. It cannot cover you against a third party — if your testing reaches\none of our providers' systems, that is between you and them.\n\n### The rules\n\n- **Do not access, modify, delete or download anyone else's data.** Use your own test account. If you can\n reach another customer's data, that is the finding — stop there and tell us. You do not need to collect\n it to prove it.\n- **Stop as soon as you have demonstrated the issue.** Do not pivot, escalate further, or establish\n persistence.\n- **No denial of service**, volumetric testing, load testing or resource exhaustion.\n- **No social engineering** of our people, our customers or our providers, and no physical attempts against\n our premises or equipment.\n- **No spam, no mass account creation, no brute forcing** of live accounts.\n- **Give us a reasonable time to fix it before you publish** — 90 days from your report, or the day the fix\n ships, whichever comes first. If a fix is taking longer, talk to us; we will not use the clock to bury a\n finding.\n- **Obey the law**, and do not do anything that would harm the Service or the people who use it.\n\nReports produced by running an automated scanner against the live Service are not in scope, and running one\nis itself a breach of the testing section above.\n\n### Usually out of scope\n\nMissing security headers or best-practice settings with no demonstrated impact; self-inflicted\ncross-site scripting; missing rate limits without a working attack; outdated version banners without a\nworking exploit; results from a vulnerability scanner submitted without validation; social engineering;\nand issues in a third-party service rather than ours — report those to the provider that runs it.\n\nSend it anyway if you are unsure. We would rather triage a duplicate than miss a real one.\n\n### There is no paid bounty\n\n**We do not run a paid bug bounty programme and we do not pay for vulnerability reports.** We would rather\nsay that plainly at the start than have you invest a weekend on the assumption that we do. What you get is\na fast, straight response from the people who can fix it, and public credit if you want it.\n\n### Security issues in our open-source code\n\nWhere our software is published as open source, report a security issue in the code to the same address, in\nprivate. **Please do not open a public issue or a pull request that describes the vulnerability** — that\npublishes a working exploit to everyone running the code, including people who cannot patch quickly.\n\n## Shared responsibility — who secures what\n\nSecurity of a hosted service is split between the provider and the customer, and **most real incidents\nhappen on the customer's side of that line**: a shared password, a departed employee who kept access, an\nintegration token that outlived its purpose, a laptop left on a train. That is not a criticism. It is the\nreason this section exists and is specific.\n\n### What we are responsible for\n\n- The physical hardware, the facilities and the virtualisation platform.\n- The container platform, the operating systems and the runtime the application runs on.\n- Patching and updating everything in the list above, and the application itself.\n- Encryption in transit and at rest, and the management of our keys.\n- Network security, the edge, and keeping the origin off the public internet.\n- Availability, redundancy, backups of the Service, and restoring it after a failure.\n- Isolation between customers.\n- The security of our own code, our build pipeline and our deployment process.\n- Our own staff: their access, their devices, their confidentiality obligations.\n- The providers we engage, and the contracts that bind them.\n- Detecting, responding to and notifying you of incidents in our part.\n\n### What you are responsible for\n\n**Who gets in, and who stays in.** Your users, their roles and their permissions. Inviting people, removing\nthem the day they leave, reviewing periodically who still has access, and not sharing accounts. Choosing\nstrong, unique passwords and turning on multi-factor authentication where the product offers it. If your\norganisation uses single sign-on, the security of your identity provider is yours.\n\n**How you configure the product.** Every switch we give you is a decision you are making. Defaults are\nchosen to be safe, not to be right for you, and changing one is a deliberate act with consequences you own.\n\n**Your monitoring configuration.** For products that can capture workforce activity, **you decide which\ncapture features are switched on, how often they run, and who can see the result.** We supply the controls;\nwhich of them is lawful and proportionate in your organisation, in your jurisdiction, is your call. The\nnext block is about that in more detail, because it is the sharpest edge in this document.\n\n**Your retention settings.** Where the product lets you set how long data is kept, that period is yours.\nKeeping data longer than you need it is a risk you are choosing to hold, and it is one of the first things\na supervisory authority asks about.\n\n**Your integrations and your tokens.** Every third-party service you connect, every API key you issue,\nevery webhook you point somewhere and every access token you grant to an application or an automated agent.\nApproving an integration is you deciding to send your data somewhere else. Scope those grants narrowly,\nreview them, and revoke them when a person leaves or a project ends — a token issued by someone who has\nleft your organisation still works until you revoke it.\n\n**Your devices and your endpoints.** The laptops, phones and desktops your people use to reach the Service,\nincluding any desktop or mobile application we publish. Full-disk encryption, screen locks, patching,\nmalware protection and what happens when a device is lost are yours. We cannot protect an account from a\ncompromised machine that is already signed into it.\n\n**Your data, and what you put in it.** What you choose to upload, whether it belongs in this product at\nall, and whether you have the right to process it. Also your own copies — see the backups section.\n\n**Telling your own people.** Your privacy notice to your staff, customers and site visitors is yours to\nwrite. Ours describes what we do; it is not your notice, and our cookie banner is not your consent\nmechanism.\n\n### Where you use the Service to monitor people\n\nSome of our products capture information about how an organisation's own personnel work — which can include\nscreenshots, application and website usage, activity rates derived from keyboard and mouse events, and,\nwhere you switch them on, webcam stills, audio or screen recordings.\n\nFor all of that data, **you are the controller and we are your processor.** You decide why it is collected\nand how; we process it on your instructions. That split has consequences you should read carefully:\n\n- **The lawful basis is yours to establish.** Consent is generally not a valid basis in an employment\n relationship because of the imbalance of power between employer and employee, so a signed form is\n unlikely to be the answer you need.\n- **The notice to the monitored person is yours to give**, in advance, covering each type of capture that\n is switched on.\n- **The data protection impact assessment is yours to complete** where one is required, which it commonly\n is for systematic monitoring of employees. We will help: ask privacy@ever.co for the information you\n need about data categories, capture frequencies, retention, storage locations, sub-processors and\n transfer mechanisms, and we will provide it.\n- **Consultation with a works council, trade union or employee representative body is yours to run** where\n local law requires it — which it does in several European countries, before the system is introduced, not\n after.\n- **Some capture types are restricted or prohibited in some countries.** Continuous screen capture in\n particular is treated very differently across Europe.\n\n**We make no claim that any monitoring feature is lawful for you to use, anywhere.** A feature existing in\nthe product is not advice that you may switch it on, and our providing it is not our assessment of your\nlegal position. Where you enable one, you confirm to us that you have done the work above — the Data\nProcessing Addendum and the [Terms of Service](https://githands.com/tos) put that in binding terms, including\nwho answers for it if it turns out not to have been done.\n\n### Where you run the software yourself\n\nIf you deploy our open-source software on your own infrastructure, **everything above is yours** — the\nservers, the database, the storage, the backups, the encryption, the patching, the access control and the\nincident response. You are the controller of the data in it. We have no access to it, we are not your\nprocessor for it, and no commitment on this page applies to it.\n\n## How to reach us about security\n\n- **A vulnerability you have found** — abuse@ever.co. Read the responsible disclosure section first; it\n tells you what protection you have and what rules go with it.\n- **An incident, a compromised account, or something you can see that you should not be able to see** —\n abuse@ever.co, immediately. Copy privacy@ever.co if personal data is involved.\n- **A security questionnaire, a due diligence request, a request for the Data Processing Addendum, evidence\n of a restore test, or permission to test the Service** — legal@ever.co.\n- **A question about personal data, a data subject request, or a breach affecting you** —\n privacy@ever.co.\n\nAll of these addresses are monitored by the people who operate the Service. By post: **Ever Technologies LTD**,\nMladost 2, bl. 211, ent. A, Sofia 1799, Bulgaria — registered in Bulgaria under company number 204599535. We\ncorrespond in English.\n\nYou may also complain to a data protection supervisory authority. Ours is the\nCommission for Personal Data Protection (Комисия за защита на личните данни), the CPDP, at\n[https://www.cpdp.bg/](https://www.cpdp.bg/). You may instead complain\nto the authority for the country where you live or work.\n\nThis document is version 1.0.2 of the Security Overview for\n[githands.com](https://githands.com), in force from 2026-08-02. It describes the measures in\nplace at that date. Earlier versions, with the dates they applied, are at\n[https://githands.com/security](https://githands.com/security)."
420
+ "markdown": "## Security overview\r\n\r\nThis page describes how we protect GitHands Cloud and the data you keep in it. It is written to be\r\nchecked rather than admired: every measure below is something we actually run, and where we do not do\r\nsomething that a page like this usually claims, **we say so instead of leaving a gap you have to guess at**.\r\n\r\nThe short version:\r\n\r\n- We run the Service on **hardware we own and operate ourselves**, not as a tenant in someone else's cloud.\r\n The database, the object storage and the secrets store are ours. That is why the list of providers with\r\n access to your data is short.\r\n- Traffic is encrypted in transit, storage is encrypted at rest, and off-site backup copies are encrypted\r\n **before they leave our network**, so the provider holding them holds ciphertext.\r\n- Production changes go through version control and review. Hand edits on live systems are the exception,\r\n not the routine.\r\n- **We hold no security certification, and we do not claim one.** The next section says exactly what we do\r\n instead.\r\n\r\n### What this page covers\r\n\r\nIt covers the hosted Service we operate at [githands.com](https://githands.com) — the application, its\r\ninterfaces, and the infrastructure underneath them.\r\n\r\nIt does **not** cover two things that are easy to confuse with it:\r\n\r\n- **A copy of the software you run yourself.** Where our software is published as open source and you deploy\r\n it on your own infrastructure, you are the operator. You choose its hosting, its storage and every other\r\n component, we have no access to it, and nothing on this page describes it.\r\n- **Your own systems and your own configuration.** Which of your people have access, how you configure the\r\n product, what you connect it to and how you secure the devices your team uses are yours. The shared\r\n responsibility section near the end of this page sets out that split in detail, because most real\r\n incidents happen on one side of that line rather than in the middle of it.\r\n\r\n### What this page is, legally\r\n\r\nThis is a factual description of the measures in place at the version and date shown at the end. It is\r\n**not a warranty, and it does not create a service level commitment.** Our contractual security obligations\r\nare in our Data Processing Addendum and in the [Terms of Service](https://githands.com/tos), and those are the\r\ndocuments that bind us.\r\n\r\nWe change these measures as the systems and the threats change. **We will not make a change that materially\r\nreduces the protection described here**, and when the description changes we republish this page with a new\r\nversion and date.\r\n\r\n### One thing we will not tell you\r\n\r\n**No system is perfectly secure, and a vendor who tells you otherwise is selling something.** Everything\r\nbelow reduces risk; none of it eliminates it. What we can commit to is that we take these measures\r\nseriously, that we test the ones that can be tested, and that when something goes wrong we tell you rather\r\nthan manage the news.\n\n## Certifications — what we hold, and what we do not\r\n\r\n**We do not hold a SOC 2 report of any type. We are not certified to ISO/IEC 27001. We hold no other\r\nsecurity certification or attestation, and we do not claim one.**\r\n\r\nWe are stating that first, in plain words, because a security page that leaves the question open is doing\r\nso on purpose. If a proposal, a sales conversation, a questionnaire response, a directory listing or a\r\nthird-party summary has told you that we are SOC 2 or ISO 27001 certified, **it is wrong** — please send it\r\nto legal@ever.co so we can get it corrected at the source.\r\n\r\n### What we do instead\r\n\r\nWe use two published control sets as a working checklist rather than as a badge:\r\n\r\n- **The control areas of ISO/IEC 27001 Annex A** — access control, cryptography, operations security,\r\n supplier relationships, incident management, continuity — as the structure for reviewing our own\r\n practices. This page is organised broadly along those lines.\r\n- **Article 32 GDPR**, which is not optional for us and is the standard our customers can actually hold us\r\n to: technical and organisational measures appropriate to the risk, kept under review as the risk changes.\r\n\r\n**No external auditor has examined either.** These are self-assessments, and you should read them as such.\r\nAn unaudited framework alignment and a certification are different things, and we will not blur the two by\r\nwriting \"aligned with\" in a way that reads like \"certified to\".\r\n\r\n### What you can have instead of a certificate\r\n\r\nA certificate is a proxy for evidence. Where we cannot give you the proxy, we will give you the evidence:\r\n\r\n- **This page**, which is specific enough to be checked against what we actually do.\r\n- **The security measures schedule in our Data Processing Addendum**, which is contractual rather than\r\n descriptive — if a measure matters to you, that is where to pin it down.\r\n- **A completed security questionnaire.** Send yours to legal@ever.co. We answer them, we answer them\r\n honestly, and where the honest answer is \"no\" you will get \"no\" rather than a paragraph that avoids the\r\n question.\r\n- **Evidence on request** for specific points — a description of the architecture, the outcome of our most\r\n recent restore test, the current sub-processor list, the transfer mechanism for a named provider.\r\n- **A right to audit**, on the terms set out in the Data Processing Addendum.\r\n\r\n### If that changes\r\n\r\nIf we obtain a certification or an independent report, this page will say which one, who issued it, what\r\nits scope covers and what it excludes — because the scope is the part that matters and the part most often\r\nleft out. Until that day, the sentence at the top of this section is the whole answer.\n\n## Encryption\r\n\r\n### In transit\r\n\r\nAll traffic between you and the Service travels over TLS. HTTPS is enforced and plain HTTP requests are\r\nredirected rather than served, so a request that arrives unencrypted never reaches the application with its\r\ncontents intact. We keep the accepted protocol versions and cipher suites current and retire weak ones as\r\nthe ecosystem does.\r\n\r\nTraffic between our own sites, and between our infrastructure and the edge network in front of it, runs\r\nover encrypted connections. Traffic to the providers we engage runs over TLS to their published endpoints.\r\n\r\n### At rest\r\n\r\n- **Storage holding customer content, databases and backups is encrypted at rest.** That protects against\r\n the loss, theft or improper disposal of physical media — it does not protect against a compromise of a\r\n running system that already has the data decrypted, and we would rather you knew which threat it answers.\r\n- **Off-site backup copies are encrypted before they leave our network**, with keys we hold and manage\r\n ourselves. The provider storing those copies holds ciphertext and nothing else: it cannot read them, and\r\n neither can anyone who obtains them from it.\r\n\r\n### At the application layer\r\n\r\nSome values are too sensitive to rely on storage-level encryption alone, so they are encrypted again inside\r\nthe application before they are written:\r\n\r\n- **Credentials, access tokens and API keys** — including tokens you grant us for a third-party integration.\r\n- **Secrets used to operate the Service** — held in a secrets manager we run ourselves, injected into\r\n workloads at runtime, and **never committed to source control**.\r\n\r\n**Passwords are never stored in a form we can read.** They are put through a modern one-way hashing\r\nfunction with a per-password salt. We cannot recover your password, tell you what it is, or send it back to\r\nyou — a reset is the only route, and any service that offers you the alternative is storing it wrongly.\r\n\r\nWhere a specific product stores a particular credential differently from the description above, its annex\r\nto this page says so and says why. **We would rather publish an awkward exception than a tidy sentence that\r\nis not true everywhere.**\r\n\r\n### Keys\r\n\r\nEncryption keys are held in our self-hosted secrets manager, separately from the data they protect, and are\r\naccessible only to the systems and people that need them. Keys are rotated when a person with access leaves,\r\nwhen we have reason to think a key has been exposed, and on a periodic schedule. **The keys for off-site\r\nbackups are held independently of the infrastructure they protect**, so that losing a site does not also\r\nlose the ability to decrypt its backups.\n\n## Access control\r\n\r\n### Your access to your own data\r\n\r\nAccess inside your workspace is yours to configure. The Service gives you named user accounts, roles and\r\npermissions, workspace administrators who can invite and remove people, and session controls. Where a\r\nproduct supports single sign-on, multi-factor authentication or IP restrictions, its annex says so and its\r\ndocumentation says how to switch them on.\r\n\r\n**Turning those controls on, and reviewing who still needs access, is your job — see the shared\r\nresponsibility section.** The most common way an account is compromised is not a flaw in our systems; it is\r\na shared password, a former employee who was never removed, or an integration token that outlived its\r\npurpose.\r\n\r\n### Our access to production systems\r\n\r\n- **Named individual accounts, never shared logins.** Every action in production is attributable to a\r\n person.\r\n- **Multi-factor authentication** on the systems and administrative accounts that support it.\r\n- **Least privilege.** Access is scoped to the narrowest role that lets the work happen, and elevated\r\n access is granted for a purpose rather than held permanently by default.\r\n- **Changes go through Git.** The configuration of the live environment is held in version control and\r\n applied automatically, so a change is reviewable, attributable and reversible. Direct hands-on changes to\r\n a running system are an exception for incidents and hands-on maintenance, not the daily routine — and the\r\n automation reverts undeclared edits rather than leaving them in place.\r\n- **Removal on departure or role change.** Access is revoked when someone leaves or no longer needs it, and\r\n credentials they held are rotated. Access is reviewed periodically as well, because revocation lists rot.\r\n\r\n### Access to your content\r\n\r\nWe access the content in your workspace only where it is necessary to operate, secure or support the\r\nService — for example to investigate a fault you have reported, or to respond to a security incident. **We\r\ndo not browse customer content, and we do not use it to build products, train models or produce insights\r\nfor anyone else.**\r\n\r\nSupport access is granted for the task in hand and logged. Where a product offers a mechanism for you to\r\ngrant or revoke support access yourself, its annex describes it.\r\n\r\n### Being honest about the limits\r\n\r\n**We are a small team, and several of us hold administrative access because operating this infrastructure\r\nrequires it.** A large organisation would separate the person who approves a change from the person who\r\nmakes it; at our size that separation is partial, and we are not going to describe it as complete. What we\r\ndo instead is make every change visible: production state comes from version control, actions are\r\nattributable to named accounts, and administrative events are logged and monitored.\n\n## Infrastructure and hosting\r\n\r\n### We run our own infrastructure\r\n\r\nThe Service does not run in a rented public cloud account. It runs on **physical servers we own, in\r\nfacilities we control, inside the European Union**, on a virtualisation and container platform we operate\r\nourselves. The database, the object storage, the message and cache layers, the secrets store and the\r\ndeployment system are all components we run — not managed services bought from a third party.\r\n\r\nThat is an unusual answer in this market, and it has a direct consequence you can verify: **the list of\r\nproviders with access to your data is short.** Most of what would otherwise be a sub-processor is simply a\r\nservice we operate.\r\n\r\nIn outline, and deliberately without the detail that would help someone attack it:\r\n\r\n- **Hardware and virtualisation we own**, clustered so that losing a machine does not lose a service.\r\n- **A container orchestration platform** running the applications, with the control plane replicated across\r\n multiple machines.\r\n- **PostgreSQL**, run by an operator that handles replication, failover and continuous archiving of the\r\n write-ahead log.\r\n- **S3-compatible object storage** we host, for files and media, with data replicated across nodes.\r\n- **A secrets manager we host**, from which credentials are injected into workloads at runtime.\r\n- **Declarative configuration held in Git** and applied automatically, so the live environment matches a\r\n reviewed commit rather than an accumulation of manual changes.\r\n\r\nA content delivery and security network sits in front of all of it, and our source code, build pipeline and\r\ncontainer registry are hosted with a third-party provider. Those are genuine third parties even though the\r\ninfrastructure behind them is ours, and they are named on our\r\n[sub-processor page](https://githands.com/subprocessors).\r\n\r\n### Redundancy and failure domains\r\n\r\nServices that can run on more than one machine run on more than one machine. Databases are replicated,\r\nobject storage is replicated across nodes, and the systems that hold backups run on **a physically separate\r\ncluster from the systems they protect** — so a failure that takes out the live environment does not take\r\nits backups with it.\r\n\r\n**This is not the same as multi-region redundancy, and we will not describe it as such.** We can lose\r\nmachines without losing data or availability. A whole-site failure means recovering from off-site copies,\r\nwhich is a matter of hours rather than seconds. The continuity section is specific about what that means.\r\n\r\n### Separation between customers\r\n\r\nIsolation between customers is **logical, not physical**, unless you have separately contracted for a\r\ndedicated deployment. Depending on the product that means a separate database per customer, a separate\r\nnamespace and deployment per customer site, or tenant scoping enforced in the application and the data\r\nmodel. **Your product's annex states which model applies to you** — it is a question worth asking of any\r\nvendor, and worth a straight answer.\r\n\r\n### Separation between environments\r\n\r\nDevelopment, staging and production are separate environments with separate credentials and separate data\r\nstores. Production credentials are not available in development or staging, and **we do not copy production\r\ndata into a development environment**. Test data is synthetic or anonymised.\r\n\r\n### Third-party managed services, where they are used\r\n\r\nA small number of products use a third-party managed database, managed storage or hosted platform for a\r\nspecific function rather than our own. Where that is the case it is named in that product's annex and it\r\nappears in the always-engaged tier of the sub-processor list. **We do not describe the estate as fully\r\nself-hosted where a product's data actually sits somewhere else.**\n\n## Network security\r\n\r\n### At the edge\r\n\r\nEvery request to the Service passes through a content delivery and security network before it reaches us.\r\nThat layer terminates TLS, absorbs volumetric denial-of-service traffic, applies a web application firewall\r\nand filters automated abuse. It also means the address of our origin infrastructure is not the address your\r\nbrowser connects to.\r\n\r\n### The origin is not exposed\r\n\r\n**Our application servers do not publish inbound ports to the public internet.** Traffic reaches them\r\nthrough outbound-initiated tunnels established from inside our network to the edge, so there is no\r\npublic listener to scan, and no direct path from the internet to an origin host.\r\n\r\n### Inside the network\r\n\r\n- Databases, object storage, the secrets manager and internal services sit on **segmented private networks\r\n and are not reachable from the internet**. They accept connections only from the workloads that need them.\r\n- **Administrative interfaces are not published publicly.** Reaching them requires being on the internal\r\n network, through a controlled route with authenticated access.\r\n- Outbound connectivity from workloads is limited to what the workload actually needs, so a compromised\r\n component has fewer places to send data and fewer places to fetch a second stage from.\r\n\r\n### Abuse and automation\r\n\r\nSign-up, sign-in, password reset and contact endpoints are protected against automated abuse by rate\r\nlimiting and by a bot-detection challenge. **That challenge is strictly necessary to keep the Service\r\nworking and available** — it protects accounts from credential-stuffing and forms from bulk abuse — which\r\nis why it runs without a consent prompt, as explained in our [Cookie Policy](https://githands.com/cookies). It\r\nis not used for analytics, profiling or advertising.\r\n\r\n### What we monitor for\r\n\r\nUnusual authentication patterns, unexpected administrative activity, sudden changes in traffic shape and\r\nfailures of the components above raise alerts that a person sees. The logging and monitoring section\r\ndescribes what is recorded and for how long.\n\n## Application security\r\n\r\n### How a change reaches production\r\n\r\nEvery change to the Service starts as a change to source code in version control and follows the same\r\nroute:\r\n\r\n1. **A pull request**, reviewed by someone other than its author on the branches that matter.\r\n2. **Automated checks** — build, tests, static analysis and dependency scanning — which must pass.\r\n3. **Promotion through environments.** Changes move development to staging to production, in that order.\r\n Production is deployed from the release branch and from nothing else.\r\n4. **A container image built by the pipeline** from a reviewed commit, identified by its digest, deployed\r\n by the same automation that holds the environment's configuration in Git.\r\n\r\n**Nobody edits production by hand as a way of shipping a change.** The deployment system reconciles the\r\nlive environment against the version-controlled definition, so an undeclared edit is reverted rather than\r\nquietly kept.\r\n\r\n### Automated checks on every change\r\n\r\n- **Dependency scanning**, with alerts for known vulnerabilities in the packages we use, and updates\r\n applied on a defined schedule with a faster path for serious issues.\r\n- **Static analysis** of our source code, run on each change, flagging the classes of defect it can find\r\n before the code ships.\r\n- **Secret scanning** across our repositories, with pushes blocked where a credential is detected. Where a\r\n credential has been exposed, we treat rotating it at the provider as the fix — deleting it from the code\r\n is not the fix.\r\n- **Automated tests**, which have to pass before a change can be merged or released.\r\n\r\nDependencies are installed through an internal package cache we run, which gives us a consistent, auditable\r\nview of what actually went into a build.\r\n\r\n### In the application itself\r\n\r\n- Database access goes through parameterised queries and a mapping layer rather than assembled strings.\r\n- Output is encoded for its context, and state-changing requests are protected against cross-site request\r\n forgery.\r\n- Authentication, session handling and password storage use established libraries rather than anything we\r\n invented; passwords are hashed as described in the encryption section.\r\n- Authorisation is enforced on the server for every request. A hidden control in the interface is not an\r\n access control.\r\n- Input is validated at the boundary, and uploads are constrained by type and size.\r\n- Browser-facing responses carry security headers, including a content security policy on the surfaces\r\n where one can be applied without breaking the product.\r\n\r\n### The honest limit\r\n\r\n**Automated tooling finds categories of defect, not all defects.** Review catches more, and neither catches\r\neverything. That is why the sections on logging, incident response and responsible disclosure exist: they\r\nare what we rely on for the bugs that get through, and something always gets through.\n\n## Testing\r\n\r\n### What we test, and how often\r\n\r\n- **Automated test suites** run on every change, in the pipeline, before anything can be merged or\r\n released.\r\n- **Static analysis and dependency scanning** run on every change as well, as described in the previous\r\n section.\r\n- **Restore tests.** We restore backups into a scratch environment and check the result against the source\r\n rather than assuming a backup that completed is a backup that works. This is the test most often skipped\r\n in the industry and the one most often needed.\r\n- **Manual review** of changes that touch authentication, authorisation, payment flows, data export or the\r\n handling of credentials — the areas where an automated check is least likely to notice that the logic is\r\n wrong.\r\n- **Configuration review** of the version-controlled definition of the live environment, which is where a\r\n surprising share of real exposure lives.\r\n\r\n### Penetration testing — what we do not have\r\n\r\n**We have not commissioned an independent penetration test of the hosted Service, and we do not claim one.\r\nThere is no test report for us to send you, and any summary suggesting otherwise is not ours.**\r\n\r\nWe would rather write that sentence than the usual formulation about \"regular security assessments\" that\r\nturns out on inspection to mean an automated scanner. When we do commission a test, this page will say who\r\nperformed it, when, what was in scope and what was excluded.\r\n\r\n### Testing the Service yourself\r\n\r\n**You may not run security testing against the Service without our written agreement** — automated\r\nscanning, load testing, fuzzing and exploitation attempts included. From our side an unannounced scan is\r\nindistinguishable from an attack, and it will be handled as one.\r\n\r\nIf you want to test, we will usually say yes. Write to legal@ever.co and we will agree the scope, the\r\nwindow, the accounts you use and how findings are reported. The conditions are the obvious ones: your own\r\ntest data only, no other customer's data touched, nothing that degrades the Service for anyone else, and\r\nfindings to us before they go anywhere else.\r\n\r\nIf you have already found something without meaning to, do not test further — go straight to the\r\nresponsible disclosure section, which is written for exactly that situation and offers you protection for\r\nit.\n\n## Logging and monitoring\r\n\r\n### What we record\r\n\r\n- **Authentication events** — sign-in, sign-out, failed attempts, password and multi-factor changes.\r\n- **Administrative and security-relevant events** — permission changes, invitations and removals, token\r\n creation and revocation, changes to workspace settings, access to production systems.\r\n- **Application programming interface access** — the request, its source, its outcome and its timing.\r\n- **Infrastructure and platform events** — deployments, configuration changes, node and service state.\r\n- **Application errors and diagnostics** — enough to work out what broke.\r\n\r\n### What we try not to record\r\n\r\nWe do not deliberately log passwords, session tokens, API keys or payment card numbers, and where such a\r\nvalue would otherwise appear in a log or an error report it is redacted. **We will not claim this is\r\nperfect** — a value can end up in a stack trace nobody predicted. When we find one, we treat it as an\r\nexposure: the credential is rotated and the log entry removed.\r\n\r\nWhere a product sends error reports to a third-party error-tracking service, that service is named in the\r\nproduct's annex and on the [sub-processor page](https://githands.com/subprocessors), because an error report can\r\ncarry an identifier for the person who hit the error.\r\n\r\n### How long we keep it\r\n\r\n| Log type | Retention |\r\n|---|---|\r\n| Security, access and audit logs | **12 months** |\r\n| Application and error logs, crash reports, diagnostics | **90 days** |\r\n\r\nThese are the same periods published in the retention table of our\r\n[Privacy Policy](https://githands.com/privacy), and if the two ever disagree the Privacy Policy governs.\r\n\r\nAccess to logs is restricted to the people whose job needs it, and log access is itself an administrative\r\nevent that gets recorded.\r\n\r\n### Monitoring and alerting\r\n\r\nMetrics from the platform, the databases, the storage layer and the applications feed a monitoring system\r\nthat raises alerts when something is wrong — a service failing, a certificate close to expiry, a backup\r\nthat did not complete, a disk filling, an unusual pattern of authentication failures. **Alerts go to people,\r\nnot to a dashboard nobody opens.**\r\n\r\nWe also run an **independent external check** that watches the Service and the monitoring system itself\r\nfrom outside our network, and raises an alarm if either goes quiet. Monitoring that cannot survive the\r\nfailure it is meant to detect is decoration, and a failed backup that alerts nobody is the standard way a\r\nrecovery turns out to be impossible.\n\n## Incident response\r\n\r\n### How we handle an incident\r\n\r\n1. **Detect and record.** An alert, a customer report or a disclosure from a researcher opens an incident,\r\n with one named person responsible for running it.\r\n2. **Assess and classify.** What happened, what is affected, whether personal data is involved, whose data,\r\n and how bad it could get. Severity drives everything that follows.\r\n3. **Contain.** Cut the access, revoke the credential, isolate the component, block the route — stopping it\r\n comes before understanding it.\r\n4. **Eradicate and recover.** Remove the cause, restore service, restore data from backup where that is\r\n what is needed, and verify the result rather than assuming it.\r\n5. **Notify.** Under the timescales below.\r\n6. **Review afterwards.** A written review of what happened, why, what we changed, and what would have\r\n caught it sooner.\r\n\r\n### Telling you about it\r\n\r\n**Where we are the controller** — for the data described in our [Privacy Policy](https://githands.com/privacy) as\r\nours — and there is a personal data breach that is likely to result in a risk to people:\r\n\r\n- We notify the CPDP **within 72 hours of becoming aware of it**, as\r\n Article 33 GDPR requires, and where we cannot give complete information in that window we give what we\r\n have and follow up.\r\n- We notify the affected people **without undue delay** where the breach is likely to result in a high risk\r\n to them.\r\n\r\n**Where we are your processor** — everything inside your workspace, including data your organisation\r\ncollects about its own personnel — you are the controller and the notification duty to the authority is\r\nyours, so ours is to you:\r\n\r\n- We notify you **without undue delay after we become aware, and in any event within 72 hours.**\r\n- We give you what Article 33(3) requires you to have: the nature of the breach, the categories and\r\n approximate number of people and records involved, the likely consequences, the measures we have taken or\r\n propose, and a contact point who can answer follow-up questions.\r\n- We assist you with your own notifications and with any questions from your supervisory authority. Our\r\n Data Processing Addendum sets that out contractually.\r\n\r\nWe notify you at the administrative contact addresses on your account. **Keeping those current is worth\r\nfive minutes of your time**, because a notification that reaches a mailbox nobody reads has been made and\r\nnot received.\r\n\r\n### What we will not do\r\n\r\n- **We will not sit on a breach** to work out how it looks, or wait until we have a complete picture before\r\n telling you something is wrong.\r\n- **We will not require you to sign anything** — a non-disclosure agreement, a waiver, an amended contract —\r\n as a condition of being told.\r\n- **We will not describe an incident as something else.** If data was accessed, the notice says data was\r\n accessed.\r\n\r\n### Reporting an incident to us\r\n\r\nIf you think an account has been compromised, credentials have leaked, or you have seen something in the\r\nService you should not be able to see, tell us immediately at abuse@ever.co. If personal data is\r\ninvolved, copy privacy@ever.co. Both addresses are monitored and reach the people who operate the\r\nService.\r\n\r\nIf you have found a vulnerability rather than an incident in progress, the responsible disclosure section\r\nbelow is the route, and it offers you protection that this section does not.\n\n## Continuity and backups\r\n\r\n### Three tiers, and why there are three\r\n\r\nBackups fail in three different ways — the copy is too old, the copy is on the machine that died, or the\r\ncopy is in the building that burned. So there are three tiers, each answering a different failure:\r\n\r\n- **Tier 1 — a fast local copy.** Recent database backups are staged on storage separate from the live\r\n database, on a different machine. This is the tier that answers \"somebody dropped the wrong table twenty\r\n minutes ago\" and it restores quickly because the data never left the site.\r\n- **Tier 2 — a durable on-site copy.** Everything is copied into a **private, backup-only object store that\r\n runs on a physically separate cluster from the systems it protects**, with its own credentials, reachable\r\n only from inside our network and never from the internet. Losing the cluster that runs the Service does\r\n not lose the cluster that holds its backups.\r\n- **Tier 3 — an off-site copy with a third party.** A further copy goes to object storage operated by an\r\n external provider, in a different place entirely, **encrypted before it leaves our network**. The\r\n provider stores ciphertext and holds no key. This is the tier that answers the loss of the whole site.\r\n\r\nDeliberately, **the credentials that run the Service are not the credentials that can delete its backups.**\r\n\r\n### What is backed up\r\n\r\n- **Databases** — continuous archiving of the write-ahead log for point-in-time recovery, plus periodic\r\n full backups and separate logical dumps. Two mechanisms, because they fail differently.\r\n- **Object storage** — the files and media held for you, mirrored to the backup tiers.\r\n- **Cluster and platform state**, including the configuration needed to rebuild the control plane.\r\n- **The secrets store**, snapshotted on a short cycle.\r\n- **The version-controlled configuration** that defines the environment, which is what makes a rebuild\r\n possible rather than merely theoretical.\r\n\r\n**Not backed up, on purpose:** caches and queues, which are disposable by design and reconstructed on\r\nrestart, and build artefacts that the pipeline can regenerate. Backing those up would add cost and\r\nrestore time for nothing.\r\n\r\n### Recovery objectives\r\n\r\nThese are the objectives we design and test against. **They are internal targets, not contractual\r\ncommitments** — no service level applies unless we have separately agreed one with you in writing, as the\r\n[Terms of Service](https://githands.com/tos) says.\r\n\r\n| Scenario | Recovery point — how much data is at risk | Recovery time — how long it takes |\r\n|---|---|---|\r\n| Loss of a machine or a service | None. Replicated components carry on | Minutes, usually without anyone noticing |\r\n| Database corruption or an unwanted change | Minutes — continuous archiving supports point-in-time recovery | Hours, depending on the size of the database |\r\n| Loss of the storage holding files | Last completed replication cycle | Hours |\r\n| **Loss of the whole site** | Last completed off-site cycle | **Hours to days.** This is the honest number |\r\n\r\n**We are candid about the last row.** We are resilient to losing machines; we are not a multi-region\r\nservice that fails over silently, and we will not describe ourselves as one.\r\n\r\n### We test restores\r\n\r\nA backup nobody has restored is a guess. We restore into a scratch environment and verify the result\r\nagainst the source — record counts, integrity checks, and whether the restored system actually starts —\r\nrather than trusting a job that reported success. Restore drills are documented, and the outcome of the\r\nmost recent one is available to customers on request at legal@ever.co.\r\n\r\nMonitoring alerts on a backup that is late, failing or stale, because **a backup system that fails silently\r\nis functionally the same as no backup system**, and is discovered at the worst possible moment.\r\n\r\n### How long backups are kept, and what deletion means\r\n\r\nEach backup ages out on its own rolling cycle, within the period given in the retention table of our\r\n[Privacy Policy](https://githands.com/privacy). Backups are not individually edited to remove one record — that\r\nwould defeat what a backup is for — so a deleted record persists in backup copies until that cycle\r\ncompletes. During that window the data is not used for anything, and **we do not restore a backup in order\r\nto bring back data somebody asked us to delete.** A backup is restored to recover a system, not to\r\nresurrect a record.\r\n\r\n### Your own copies\r\n\r\nWe maintain backups for our operational continuity and for the recovery obligations we owe you. **They are\r\nnot a substitute for your own copies**, and the Service provides export so that you can keep some. If your\r\nbusiness could not tolerate the loss of the data you keep with us, export it on a schedule that reflects\r\nthat.\n\n## Retention, deletion and export\r\n\r\n### Getting your data out\r\n\r\nYou can export your data while your account is open — through the product's own export function where it\r\nhas one, through the application programming interface, or by asking us. Exports are in structured,\r\nmachine-readable formats — commonly CSV and JSON, with files delivered in their original format. Your\r\nproduct's annex says which export routes exist for it.\r\n\r\nIf you need an export we do not offer as a self-service function, write to legal@ever.co and we will\r\nproduce it. **We do not charge for a reasonable export, and we do not make leaving harder than joining.**\r\n\r\n### After your agreement ends\r\n\r\nYour data remains available for export for **30 days** after the agreement ends. After that window we\r\ndelete it. The same period, and the reasoning behind it, is in the retention table of our\r\n[Privacy Policy](https://githands.com/privacy).\r\n\r\nIf you need longer, ask before the window closes rather than after it.\r\n\r\n### Deletion on request\r\n\r\nWhere you delete something inside the Service, or ask us to delete your workspace:\r\n\r\n- **Live systems are updated straight away.** The data is gone from the running service.\r\n- **Backup copies are not individually edited.** A deleted record persists in backups until each copy ages\r\n out on its rolling cycle, within the period stated in the Privacy Policy. During that window the data is\r\n not used for anything, and we do not restore a backup to bring it back.\r\n- **Copies held by a provider we engage** are deleted on our instruction, under the Article 28 terms in our\r\n contract with them.\r\n\r\nWhere we can keep something useful without keeping anyone identifiable — aggregate usage counts, for\r\nexample — we anonymise rather than delete, and once anonymised it is no longer personal data.\r\n\r\n### Retention you control\r\n\r\nFor data inside your workspace we are your processor, and **you set the retention period, not us**. Where a\r\nproduct lets you configure how long activity data, captured media, logs or archived records are kept, that\r\nsetting is yours and its default is documented in the annex. Choosing a period that is lawful and\r\nproportionate in your jurisdiction is part of your controller responsibility — see the shared\r\nresponsibility section.\r\n\r\nOur own retention periods, for the data where we are the controller, are published in full in the\r\n[Privacy Policy](https://githands.com/privacy) as a table with actual numbers rather than the phrase \"as long as\r\nnecessary\".\r\n\r\n### Legal hold\r\n\r\nWhere we are required to preserve data for a legal claim, an investigation or a regulatory request, we keep\r\nit until that ends, including any appeal period, and we do not delete it in the meantime. That obligation\r\noverrides the periods above. It applies to us; it will sometimes apply to you as well.\r\n\r\n### Disposing of hardware\r\n\r\nBecause we own our infrastructure, we also retire our own drives. Storage is encrypted at rest, and a drive\r\nis wiped, cryptographically erased or physically destroyed before it leaves our control. **A retired drive\r\nis never sold, donated or returned intact.**\n\n## Personnel\r\n\r\nWe are a small, long-standing team, and the people who build the Service are the people who operate it.\r\nThat has a security upside — nobody is administering a system they do not understand — and a limit we set\r\nout in the access control section rather than hide.\r\n\r\n**Everyone with access to production systems or customer data:**\r\n\r\n- is bound by **written confidentiality obligations that continue after their engagement ends**, whether\r\n they are an employee or a contractor;\r\n- works on a **need-to-know basis**, with access scoped to their role and granted for a purpose rather than\r\n by default;\r\n- uses **named individual accounts with multi-factor authentication**, never a shared login;\r\n- is briefed on how we expect data to be handled as part of joining, and again when something material\r\n changes;\r\n- **loses that access when they leave or when their role changes**, with the credentials they held rotated\r\n as part of the same step.\r\n\r\nContractors and anyone working on our behalf are engaged on terms at least as strict as those above. We do\r\nnot give a provider or a contractor access to customer data because it is convenient.\r\n\r\n### Devices\r\n\r\nMachines used to access production systems or customer data run full-disk encryption, lock automatically,\r\nare kept patched, and are protected by multi-factor authentication on the accounts they hold. A lost or\r\nstolen device is treated as an incident: access is revoked and credentials are rotated before anything else\r\nis decided.\r\n\r\n### Two things we do not claim\r\n\r\n- **We do not run formal background screening as a standard step**, and we are not going to describe an\r\n informal hiring process as vetting. Where your own contract or regulatory position requires screening for\r\n named individuals working on your account, tell us at legal@ever.co and we will arrange it.\r\n- **We do not run a certified annual security awareness training programme.** Security expectations are\r\n part of onboarding and part of how we review changes, but that is a practice, not a curriculum, and the\r\n two should not be confused on a page like this.\n\n## Providers and vendor management\r\n\r\n### The first question is whether we need one at all\r\n\r\nWe run our own infrastructure, and that is a security decision as much as a cost one: **a provider we do\r\nnot engage is a provider that cannot be breached on our behalf.** Before we take on a new one we ask\r\nwhether the function can be served by something we already operate. Often it can, and the answer is no\r\nprovider.\r\n\r\n### Before we engage one\r\n\r\n- **What data would it see**, and can we send it less — or none — and still get the benefit?\r\n- **What is its security posture**: its own certifications and reports, its published architecture, its\r\n breach history, how it handles vulnerability reports.\r\n- **Where does it process**, which establishment contracts with us, and which transfer mechanism applies if\r\n processing happens outside the European Economic Area.\r\n- **Will it sign a data processing agreement** meeting Article 28 GDPR, including the Standard Contractual\r\n Clauses where they are needed? If it will not, we do not engage it for anything involving personal data.\r\n- **How do we get out** — can we export what it holds, and can we replace it without rebuilding the\r\n product around its absence?\r\n\r\n### What the contract has to contain\r\n\r\nProcessing only on our documented instructions; confidentiality binding its personnel; security measures at\r\nleast as strict as our own; controls on its own sub-processors; assistance with data subject requests and\r\nwith security, breach notification and impact assessments; notification of a breach without undue delay;\r\ndeletion or return of the data when the engagement ends; and our right to the information needed to\r\ndemonstrate compliance.\r\n\r\n**A provider may not use your data for its own purposes.** Not to improve its own products, not to train\r\nmodels, not to build advertising profiles.\r\n\r\n### While it is engaged\r\n\r\nWe review the providers we use periodically, watch for security incidents and for changes in the legal\r\nbasis for transfers to them, and remove one when the feature it served is gone. A provider that is no\r\nlonger used does not stay on the list as decoration, and a provider that has been quietly replaced is a\r\nchange we announce.\r\n\r\n### How this appears to you\r\n\r\nEvery provider that processes personal data for the hosted Service is published on our\r\n[sub-processor page](https://githands.com/subprocessors), in three tiers: those always engaged, those engaged\r\nonly when you enable a specific feature, and those that appear only in a deployment you run yourself. That\r\npage also gives the advance notice period before we add or replace one, and the procedure for objecting.\n\n## Annex: GitHands\r\n\r\nEverything above describes how we protect a customer's data. **Most of the personal data in this Service\r\nbelongs to people who are not customers** — developers enumerated from public platforms, who have no account,\r\nno password to choose, no multi-factor setting to switch on and no way to check on us. They cannot take a\r\nsingle one of the precautions the shared-responsibility section asks of you.\r\n\r\n**That moves the whole burden onto our side of the line**, and this annex is about the parts of it that are\r\nspecific to this product.\r\n\r\n### The source-platform credential you connect\r\n\r\nThe Service reads GitHub, GitLab and similar platforms under a credential **you** connect, in your own name.\r\nIt is the most sensitive thing you hand us, so here is exactly what happens to it.\r\n\r\n**What it is used for.** Reading public data through the platform's own official public API: profiles,\r\nstargazer and forker lists, contributor and participant lists, organisation followers, public commit\r\nmetadata, and the dependency manifests in public repositories. **It is not used to write anything, anywhere**\r\n— no commits, no issues, no comments, no repository changes — and it is not used to read anything that is not\r\npublic. If your grant permits more than that, we still do less.\r\n\r\n**A grant can be wider than the use.** Platforms issue access in coarse blocks, so the token you create may\r\ncarry more access than the Service needs or exercises. **Review the grant at the platform, give it the\r\nnarrowest form the platform offers, and revoke it there whenever you want** — revocation takes effect on the\r\nplatform's side immediately, and the Service simply stops being able to read.\r\n\r\n**How it is stored, stated exactly.** Tokens are held in our database, on our own infrastructure, and we are\r\nin the middle of moving them to application-layer encryption — a value encrypted inside the application\r\nbefore it is written, as described in the encryption section above. **That migration is not finished. Until\r\nit is, the token is also still written to the older column that carries no application-layer encryption**, so\r\nin that path it is protected by storage-level encryption and database access control rather than by\r\napplication-layer encryption. We are completing the migration and removing the older column. **We would\r\nrather publish that sentence than a tidier one that is not yet true**, and if you would prefer to wait for\r\nthe migration before connecting a token, that is a reasonable thing to want — ask legal@ever.co where it\r\nstands.\r\n\r\n**We do not pool credentials.** Your reads run under your credential and nobody else's, and we do not use a\r\ncustomer's token to serve another customer's work.\r\n\r\n### Keys, agents and machine access\r\n\r\n- **An API key can do what your account can do.** It is a bearer credential: whoever holds it is you, as far\r\n as the Service is concerned. Treat it like a password, keep it out of source control, rotate it, and tell\r\n us at abuse@ever.co the moment one is exposed so we can revoke it.\r\n- **The command-line client, the machine interface and connected agents all act with your credentials**, and\r\n what they do is attributed to you.\r\n- **Outbound webhooks send data to an address you configure.** Whatever is at that address receives it. Point\r\n them at endpoints you control, over HTTPS, and verify what arrives.\r\n- **A discovery document describing what the Service can do is served without authentication.** It describes\r\n capabilities, not people. Reaching any data still requires a credential — **and if you find a route that\r\n returns information about anyone without one, that is a defect we want reported**, through the responsible\r\n disclosure route below.\r\n\r\n### Where non-users' data actually goes\r\n\r\nThe sub-processor list at [https://githands.com/subprocessors](https://githands.com/subprocessors) names every\r\nprovider. Four of them are worth calling out on a security page, because they hold data about people who\r\nnever chose us:\r\n\r\n- **The AI routing provider** receives the prompt for every AI request — names, usernames, email addresses,\r\n biographies, repository lists and the drafted message — and forwards it to the model provider it routes to.\r\n- **The AI observability service** stores the full prompt and the full completion for each call as a readable\r\n trace, so that same personal data sits there in plain text.\r\n- **Our error monitoring runs with server-side logging enabled**, which means console output can travel with\r\n a report and can carry fragments of whatever request failed. That is a real exposure route and not a\r\n theoretical one, which is why we say it rather than describing \"error reports\" and stopping.\r\n- **The background-job runtime executes extraction, enrichment, scoring and sending**, so it holds job\r\n payloads and run logs — and **it holds the credentials it needs to reach our database.** A provider that\r\n holds a database credential is a provider that can reach application data. It is the highest-value\r\n dependency in this product, and treating it as merely \"a job runner\" would be misleading.\r\n\r\nEach is bound by a written contract, may use the data only on our instructions, and **may not use it for its\r\nown purposes or to train models on it.**\r\n\r\n### Separation between customers\r\n\r\nCustomer projects, campaigns and records are separated logically inside shared infrastructure — a shared\r\ndatabase with tenant-scoped records, not a database per customer. **That is logical separation, not dedicated\r\ninfrastructure, and we will not describe it as more than it is.** Access is enforced in the application on\r\nevery request; the general application-security section above describes how we build and review that.\r\n\r\n### The public pages are public on purpose\r\n\r\nThe community pages name individuals and are reachable without an account. There is nothing to breach there\r\n— it is published deliberately, and it is disclosed as such in the [Privacy Policy](https://githands.com/privacy)\r\nannex, together with the route to have an entry taken down. We mention it here so nobody reports it as a\r\nmisconfiguration, and so nobody assumes the rest of the Service is equally open. **It is not: everything\r\nabout a customer's projects, campaigns and records requires authentication.**\r\n\r\n### Suppression is a security property, not just a preference\r\n\r\nWhen someone objects, we keep **a one-way hash of their address and a date, and nothing else** — we cannot\r\nread the address back out of it, and its only job is to make sure the extraction pipeline never re-adds them.\r\n\r\nThat record is integrity-critical. **Anything that lets a suppression be bypassed, overwritten, expired or\r\nevaded is a security defect in our book**, not a product limitation, and we would rather hear about it from\r\nyou than from the person who started receiving mail again.\r\n\r\nThe same goes in the other direction: **the unsubscribe and objection routes are protected against being\r\niterated**, because a route that lets somebody walk through identifiers could be used to unsubscribe another\r\ncustomer's recipients wholesale. If you find a way to do that, tell us.\r\n\r\n### What we ask of you\r\n\r\nBeyond the general shared-responsibility split:\r\n\r\n- **Connect the narrowest credential the platform will issue**, and rotate or revoke it when the person who\r\n created it leaves your organisation. A token issued by someone who has left still works until it is\r\n revoked.\r\n- **Use a sending domain you control**, with SPF, DKIM and DMARC configured. A sending path you do not\r\n control is a security problem before it is a deliverability one.\r\n- **Treat exported records as personal data about third parties**, because that is what they are. An export\r\n sitting in a spreadsheet on a laptop is the most common way this kind of data leaks, and once it has left\r\n the Service we cannot protect it or delete it for you.\r\n\r\n### If you run it yourself\r\n\r\nEverything in this document is about the Service we operate. In your own installation, the database, the\r\nstorage, the keys, the patching, the access control and the incident response are yours.\r\n\r\n**One default is worth checking before you go live.** The administrator alert address ships pointing at an\r\naddress on our domain, so until you change it, alerts about your campaign failures — which can carry\r\noperational detail about your sending — arrive with us. Set it to your own address. The\r\n[Privacy Policy](https://githands.com/privacy) annex describes the anonymous heartbeat a self-hosted installation\r\nsends and how to switch it off.\r\n\r\n### If you are in the database rather than a customer\r\n\r\n**The responsible disclosure route below is open to you.** You do not need an account, you do not need to be\r\na customer, and you do not need our permission to look at what is publicly reachable. If you find a way to\r\nget at data about other people through this Service, tell us at abuse@ever.co and we will treat it as\r\nwhat it is — the most serious kind of report this product can receive.\n\n## Responsible disclosure\r\n\r\nIf you have found a security vulnerability in the Service, we want to hear about it, and we would rather\r\nhear about it from you than from someone using it.\r\n\r\n### How to report\r\n\r\nSend it to **abuse@ever.co**. If personal data is exposed, copy privacy@ever.co.\r\n\r\n**We do not operate a separate security mailbox, and we are not going to publish one we do not monitor.**\r\nThat address reaches the people who operate the Service.\r\n\r\nA useful report contains: what the issue is, the exact URL, endpoint or component, the steps to reproduce\r\nit, what an attacker could actually do with it, and anything you need from us to prove it. Screenshots or a\r\nshort recording help. Write in English.\r\n\r\n### What we do with it\r\n\r\n- **We acknowledge your report within three business days.**\r\n- We assess and reproduce it, and tell you what we found — including if we conclude it is not a\r\n vulnerability, and why.\r\n- We keep you updated while we fix it, and tell you when the fix is live.\r\n- We are happy to **credit you publicly** when the fix has shipped, or to leave you out of it. Your choice.\r\n\r\nSerious issues jump the queue. We fix a critical vulnerability as quickly as we can build and verify a fix,\r\nand we will tell you what the timeline looks like rather than leaving you guessing.\r\n\r\n### Our commitment to you\r\n\r\n**If you follow the rules below, we will not pursue legal action against you, we will not report you to\r\nanyone, and we will not ask your employer to.** We will treat your research as authorised access for the\r\npurposes of computer misuse legislation and our own [Terms of Service](https://githands.com/tos), and we will\r\ntreat your report as a contribution rather than a threat.\r\n\r\nThis protection covers you personally. It cannot cover you against a third party — if your testing reaches\r\none of our providers' systems, that is between you and them.\r\n\r\n### The rules\r\n\r\n- **Do not access, modify, delete or download anyone else's data.** Use your own test account. If you can\r\n reach another customer's data, that is the finding — stop there and tell us. You do not need to collect\r\n it to prove it.\r\n- **Stop as soon as you have demonstrated the issue.** Do not pivot, escalate further, or establish\r\n persistence.\r\n- **No denial of service**, volumetric testing, load testing or resource exhaustion.\r\n- **No social engineering** of our people, our customers or our providers, and no physical attempts against\r\n our premises or equipment.\r\n- **No spam, no mass account creation, no brute forcing** of live accounts.\r\n- **Give us a reasonable time to fix it before you publish** — 90 days from your report, or the day the fix\r\n ships, whichever comes first. If a fix is taking longer, talk to us; we will not use the clock to bury a\r\n finding.\r\n- **Obey the law**, and do not do anything that would harm the Service or the people who use it.\r\n\r\nReports produced by running an automated scanner against the live Service are not in scope, and running one\r\nis itself a breach of the testing section above.\r\n\r\n### Usually out of scope\r\n\r\nMissing security headers or best-practice settings with no demonstrated impact; self-inflicted\r\ncross-site scripting; missing rate limits without a working attack; outdated version banners without a\r\nworking exploit; results from a vulnerability scanner submitted without validation; social engineering;\r\nand issues in a third-party service rather than ours — report those to the provider that runs it.\r\n\r\nSend it anyway if you are unsure. We would rather triage a duplicate than miss a real one.\r\n\r\n### There is no paid bounty\r\n\r\n**We do not run a paid bug bounty programme and we do not pay for vulnerability reports.** We would rather\r\nsay that plainly at the start than have you invest a weekend on the assumption that we do. What you get is\r\na fast, straight response from the people who can fix it, and public credit if you want it.\r\n\r\n### Security issues in our open-source code\r\n\r\nWhere our software is published as open source, report a security issue in the code to the same address, in\r\nprivate. **Please do not open a public issue or a pull request that describes the vulnerability** — that\r\npublishes a working exploit to everyone running the code, including people who cannot patch quickly.\n\n## Shared responsibility — who secures what\r\n\r\nSecurity of a hosted service is split between the provider and the customer, and **most real incidents\r\nhappen on the customer's side of that line**: a shared password, a departed employee who kept access, an\r\nintegration token that outlived its purpose, a laptop left on a train. That is not a criticism. It is the\r\nreason this section exists and is specific.\r\n\r\n### What we are responsible for\r\n\r\n- The physical hardware, the facilities and the virtualisation platform.\r\n- The container platform, the operating systems and the runtime the application runs on.\r\n- Patching and updating everything in the list above, and the application itself.\r\n- Encryption in transit and at rest, and the management of our keys.\r\n- Network security, the edge, and keeping the origin off the public internet.\r\n- Availability, redundancy, backups of the Service, and restoring it after a failure.\r\n- Isolation between customers.\r\n- The security of our own code, our build pipeline and our deployment process.\r\n- Our own staff: their access, their devices, their confidentiality obligations.\r\n- The providers we engage, and the contracts that bind them.\r\n- Detecting, responding to and notifying you of incidents in our part.\r\n\r\n### What you are responsible for\r\n\r\n**Who gets in, and who stays in.** Your users, their roles and their permissions. Inviting people, removing\r\nthem the day they leave, reviewing periodically who still has access, and not sharing accounts. Choosing\r\nstrong, unique passwords and turning on multi-factor authentication where the product offers it. If your\r\norganisation uses single sign-on, the security of your identity provider is yours.\r\n\r\n**How you configure the product.** Every switch we give you is a decision you are making. Defaults are\r\nchosen to be safe, not to be right for you, and changing one is a deliberate act with consequences you own.\r\n\r\n**Your monitoring configuration.** For products that can capture workforce activity, **you decide which\r\ncapture features are switched on, how often they run, and who can see the result.** We supply the controls;\r\nwhich of them is lawful and proportionate in your organisation, in your jurisdiction, is your call. The\r\nnext block is about that in more detail, because it is the sharpest edge in this document.\r\n\r\n**Your retention settings.** Where the product lets you set how long data is kept, that period is yours.\r\nKeeping data longer than you need it is a risk you are choosing to hold, and it is one of the first things\r\na supervisory authority asks about.\r\n\r\n**Your integrations and your tokens.** Every third-party service you connect, every API key you issue,\r\nevery webhook you point somewhere and every access token you grant to an application or an automated agent.\r\nApproving an integration is you deciding to send your data somewhere else. Scope those grants narrowly,\r\nreview them, and revoke them when a person leaves or a project ends — a token issued by someone who has\r\nleft your organisation still works until you revoke it.\r\n\r\n**Your devices and your endpoints.** The laptops, phones and desktops your people use to reach the Service,\r\nincluding any desktop or mobile application we publish. Full-disk encryption, screen locks, patching,\r\nmalware protection and what happens when a device is lost are yours. We cannot protect an account from a\r\ncompromised machine that is already signed into it.\r\n\r\n**Your data, and what you put in it.** What you choose to upload, whether it belongs in this product at\r\nall, and whether you have the right to process it. Also your own copies — see the backups section.\r\n\r\n**Telling your own people.** Your privacy notice to your staff, customers and site visitors is yours to\r\nwrite. Ours describes what we do; it is not your notice, and our cookie banner is not your consent\r\nmechanism.\r\n\r\n### Where you use the Service to monitor people\r\n\r\nSome of our products capture information about how an organisation's own personnel work — which can include\r\nscreenshots, application and website usage, activity rates derived from keyboard and mouse events, and,\r\nwhere you switch them on, webcam stills, audio or screen recordings.\r\n\r\nFor all of that data, **you are the controller and we are your processor.** You decide why it is collected\r\nand how; we process it on your instructions. That split has consequences you should read carefully:\r\n\r\n- **The lawful basis is yours to establish.** Consent is generally not a valid basis in an employment\r\n relationship because of the imbalance of power between employer and employee, so a signed form is\r\n unlikely to be the answer you need.\r\n- **The notice to the monitored person is yours to give**, in advance, covering each type of capture that\r\n is switched on.\r\n- **The data protection impact assessment is yours to complete** where one is required, which it commonly\r\n is for systematic monitoring of employees. We will help: ask privacy@ever.co for the information you\r\n need about data categories, capture frequencies, retention, storage locations, sub-processors and\r\n transfer mechanisms, and we will provide it.\r\n- **Consultation with a works council, trade union or employee representative body is yours to run** where\r\n local law requires it — which it does in several European countries, before the system is introduced, not\r\n after.\r\n- **Some capture types are restricted or prohibited in some countries.** Continuous screen capture in\r\n particular is treated very differently across Europe.\r\n\r\n**We make no claim that any monitoring feature is lawful for you to use, anywhere.** A feature existing in\r\nthe product is not advice that you may switch it on, and our providing it is not our assessment of your\r\nlegal position. Where you enable one, you confirm to us that you have done the work above — the Data\r\nProcessing Addendum and the [Terms of Service](https://githands.com/tos) put that in binding terms, including\r\nwho answers for it if it turns out not to have been done.\r\n\r\n### Where you run the software yourself\r\n\r\nIf you deploy our open-source software on your own infrastructure, **everything above is yours** — the\r\nservers, the database, the storage, the backups, the encryption, the patching, the access control and the\r\nincident response. You are the controller of the data in it. We have no access to it, we are not your\r\nprocessor for it, and no commitment on this page applies to it.\n\n## How to reach us about security\r\n\r\n- **A vulnerability you have found** — abuse@ever.co. Read the responsible disclosure section first; it\r\n tells you what protection you have and what rules go with it.\r\n- **An incident, a compromised account, or something you can see that you should not be able to see** —\r\n abuse@ever.co, immediately. Copy privacy@ever.co if personal data is involved.\r\n- **A security questionnaire, a due diligence request, a request for the Data Processing Addendum, evidence\r\n of a restore test, or permission to test the Service** — legal@ever.co.\r\n- **A question about personal data, a data subject request, or a breach affecting you** —\r\n privacy@ever.co.\r\n\r\nAll of these addresses are monitored by the people who operate the Service. By post: **Ever Technologies LTD**,\r\nMladost 2, bl. 211, ent. A, Sofia 1799, Bulgaria — registered in Bulgaria under company number 204599535. We\r\ncorrespond in English.\r\n\r\nYou may also complain to a data protection supervisory authority. Ours is the\r\nCommission for Personal Data Protection (Комисия за защита на личните данни), the CPDP, at\r\n[https://www.cpdp.bg/](https://www.cpdp.bg/). You may instead complain\r\nto the authority for the country where you live or work.\r\n\r\nThis document is version 1.0.2 of the Security Overview for\r\n[githands.com](https://githands.com), in force from 2026-08-02. It describes the measures in\r\nplace at that date. Earlier versions, with the dates they applied, are at\r\n[https://githands.com/security](https://githands.com/security)."
421
421
  }