Privacy policy

1. Data controller

NETBEARS TEAM SRL, VAT ID RO37261366, Corneliu Baba 8, Iași, Romania, is the data controller for the personal data we collect to run Spellwarden for you: your account, your organisations' membership and billing records, the messages you send us and technical data, as described below.

For personal data about other people inside the content your organisation submits for processing (for example names or email addresses inside a skill package, an MCP server's tool list or an eval), your organisation is the controller and we act as its processor under Article 28 GDPR. We process that data only on your organisation's instructions and only to provide the service. To receive our data processing terms, write to hello [at] spellwarden [dot] ai.

For a rule and note your organisation shares with us, we decide how they are used, as the terms' section on rules you share with us describes, and are the controller for any personal data in them.

2. What personal data we collect

• Account data: your email address, your name, and which optional emails you have turned off. Your password is stored by Amazon Cognito, our identity provider, as a salted hash: we never see or store it in plain text. If you sign in with Google, Google sends us your email address, whether it is verified, your name and your Google account identifier. If you sign in through your organisation's single sign-on, its identity provider sends us your email address and name.
• Two-factor authentication: if you turn it on, Amazon Cognito holds the secret of your authenticator app. We store only that it is on and when it was last checked.
• Organisations and memberships: the name and URL name of each organisation you create or join, your role, who invited you, when you joined, and the email addresses an invitation was sent to and accepted with. If you are removed from an organisation or leave it, the organisation keeps a record of your user id and those addresses, so that its single sign-on does not add you back without a new invitation.
• Invitations: the email address of the person invited and who invited them. The link in the invitation is stored only as a hash.
• Audit log: who did what in an organisation and when, for example who invited a member or changed a setting. It can include an invited email address. It does not record your IP address or browser details.
• Content your organisation submits: skill packages (SKILL.md files and the files that come with them), MCP server addresses, tool lists and their input schemas, the findings your members report, the policy packs you build, and the rules your managers and owners share with us with their notes, together with the findings, exports and eval results the service produces. Scans, evals and the private registry are being released in stages; this applies to each one as it becomes available.
• API keys: we store a SHA-256 hash of each key, its first characters and its last 4 characters, never the full key.
• Connected apps: if you connect an AI app to Spellwarden, we store the details of that connection and the tokens we issue to it (see connected apps).
• Billing data: for each organisation, its plan, number of members, billing period and the customer and subscription identifiers Creem gives us. Card details and the payer's email address are collected and kept by Creem, the seller of record. We never see or store a card number.
• Messages you send us: email you send to hello [at] spellwarden [dot] ai (see inbound email).
• Technical logs: for every request to the app, Amazon API Gateway records the time, method, path, status and the IP address it was received from (usually an address of our content delivery network, Amazon CloudFront). The app's own logs record your user id, the organisation id, the route and the status, never your email address. To block abuse, our rate limiter keeps the IP address of a request for about a day.
• Runtime telemetry: when your organisation sends the OpenTelemetry traces and logs of its agents and gateways, the fixed list of fields and the session ids described in runtime telemetry, and prompts, answers and tool content only while an owner has turned content storage on.
• Eval exports: the exports your organisation requests, which can hold your MCP servers' tool descriptions and, for owners and managers, recorded sessions (see eval exports); the export destinations its owner adds, which hold account, tenant and workspace identifiers and never a secret; and an audit log entry, with the user who made it, for every export and push.
• Live demo: The live demo signs you in to a shared demonstration account. Nothing you do in it is kept, and the organisation it shows is reset every night. The technical logs and the rate limiter described above apply to the demo as to the rest of the app.
• Website logs: for every request to spellwarden.ai, Amazon CloudFront records the time, the page or file requested with its query string, the status, the page that referred you, your IP address, the browser and device details your browser sends, and connection details such as the TLS version. It does not record cookies.
• Usage data, only if you allow analytics on spellwarden.ai: the pages you view, the page that referred you, your approximate location (derived from your IP address, which Google Analytics 4 does not log or store), your device and browser type, and a pseudonymous identifier kept in a first-party cookie. See the cookie policy. The app at dash.spellwarden.ai has no analytics.

3. Connected apps (MCP)

Spellwarden has an MCP server. It lets an AI app that you allow read and act on your organisation's Spellwarden data for you. You connect an app with OAuth 2.1: you sign in to Spellwarden and allow the app on a consent screen. Nothing is connected until you allow it. An AI app you connect (for example Claude) receives the data you ask it to fetch and processes it under its own terms. It is your tool, not our processor.

When you allow an app, we issue it an access token valid for 1 hour and a refresh token valid for 7 days, which is replaced each time the app uses it. A connection lasts at most 30 days. We store the authorisation code and the refresh token only as SHA-256 hashes, never in plain text. We also keep the name the app gave itself, the address it asked to be sent back to, how you signed in and when the connection was last used. You can have up to 5 connected apps at once, and you can revoke any of them in the app at any time, which stops its tokens working.

4. Inbound email

Email sent to hello [at] spellwarden [dot] ai (or to any other address at spellwarden.ai) is received by Amazon SES in Frankfurt, which scans it for spam and viruses. Each message is then screened automatically for spam before anyone reads it, first by fixed rules and, where those cannot decide, by an AI model on Amazon Bedrock in the EU. A message judged to be spam is held back and not forwarded; we review the held messages once a day and can still release one. All other mail is forwarded to the mailbox we read and answer it from. A full copy of each message, spam included, is stored, encrypted, in our AWS storage in Frankfurt and deleted automatically after 90 days, together with the record of any message held as spam.

5. Evals and Amazon Bedrock

Evals measure what skills change and how clear an MCP server's tool descriptions are. When someone in your organisation starts an eval run, we send to AI models on Amazon Bedrock: Spellwarden's own test tasks; for a skill eval, the SKILL.md files of the skills they chose; for a tool-grade run, the names, descriptions and input schemas of the server's tools; and the first model's answers, which a second model grades. All of it is content your organisation has already given us. We never send a password, an API key, a connection token or any other Spellwarden credential.

Before we send your organisation's text, and again before we store an answer, we mask text shaped like a credential, such as access keys and tokens. Masking works by recognising patterns, so it cannot guarantee that no secret or personal data is sent: do not put secrets or personal data in the skills and tool descriptions you evaluate.

The calls go from our AWS account in eu-central-1 (Frankfurt) to Bedrock's EU cross-region inference profiles, which keep them in EU regions. Amazon Bedrock's documented data protection is that prompts and answers are not used to train any model and are not shared with the model's developer. We keep the models' answers for 30 days and then delete them. We keep the scores, the reference items each answer met, the graders' short explanations (masked) and a SHA-256 hash of each answer for as long as the organisation's records are kept, and the masked skill text and tool lists sent to a run for 400 days. Eval results are informational: they do not change a skill's curation score, its admission to a private registry or the organisation's trust score.

6. Runtime telemetry

Your organisation can send the OpenTelemetry traces and logs of its AI agents and gateways to otel.spellwarden.ai with an ingest key, so that Spellwarden can show which agents, models, tools, MCP servers and skills they used. We act as your organisation's processor for this telemetry, and we process and store it in AWS eu-central-1 (Frankfurt).

We keep only a fixed list of fields from each record: what kind of step it was (an agent, a model call, a tool or MCP call, a skill, or a decision that allowed or denied a tool call) and the name the sender gave it; the ids that link the steps of a trace and a tool call to its result; the names of the agent, the model and its provider, the tool, the MCP method, the MCP server and the skill; the host name or IP address and port of the MCP server or other service a step called or served, its URL and where a skill was loaded from, each without user names, query strings or fragments and with token-like path parts replaced; the name, version, environment and SDK language of the service that sent it; when it happened, how long it took, whether it failed and with which error type, its token counts, and for a decision its outcome, source, category and reason code. Every other attribute is dropped when it arrives. The attributes that identify a person, an account or a client, such as user ids, user email addresses, account and organisation ids and client IP addresses, are never stored. Whatever your tools write into one of the kept fields, such as an agent or service name, is kept as they sent it.

Session ids are kept as your organisation's tools sent them, so that you can find a session in your own systems, except that a session id shaped like an email address is replaced by a keyed hash, which Spellwarden does not reverse.

Prompts, answers and tool content are dropped when they arrive unless an owner of the organisation turns on content storage. While it is on, that content is kept after masking text shaped like a credential, kept for 30 days and shown only to the organisation's owners and managers, and each opening is recorded in the audit log. Stored content holds whatever personal data the prompts, answers and tool results themselves contain, such as names or email addresses: only text shaped like a credential is masked. When an owner turns content storage off, stored content is deleted within two hours. When an organisation is deleted, its stored content is deleted within two hours after its one-day restore window ends.

With content storage on, someone in your organisation can start a session eval: up to 50 recorded sessions are sent, masked, to the judge model on Amazon Bedrock in the EU, as described for evals above. Their transcripts are deleted when the run ends. The judge's explanations count as content: they are shown only to owners and managers, and deleted after 30 days, or within two hours when content storage is turned off. Session details are kept for 30 days. Daily totals, which keep the names, addresses, URLs and skill sources of what was used, are kept for 400 days. Runtime results are informational: they do not change a score, a curation or a registry admission.

7. Eval exports

Your organisation can export Spellwarden's eval rubrics as files for AWS AgentCore Evaluations, Microsoft Foundry and Outcomes in Claude Managed Agents, and download them or push them into an account of its own on one of those platforms. An export can hold the rubrics; Spellwarden's 20 sample infrastructure tasks with their reference answers; the names, descriptions and input schemas of one of your MCP servers' tools, masked, with the grades Spellwarden gave them; and, for owners and managers only, the recorded sessions of a finished session eval, which are content (see runtime telemetry above): masked prompts, answers and tool results with the judge's verdicts and explanations.

We keep each export for 30 days so that it can be downloaded again. An export of recorded sessions is stored with the other content: it is deleted after 30 days, and within two hours when content storage is turned off. Every export and every push is recorded in the audit log.

Pushes send an export only to the accounts your organisation's owner adds as export destinations: an AWS role, a Microsoft Entra app registration or a Claude service account. Pushes use no stored secret: for each push, Spellwarden obtains a short-lived credential that the destination accepts, uses it for that push only and keeps none. The request for that credential carries Spellwarden's AWS account and your organisation's and the destination's identifiers, so a destination accepts pushes only from the organisation that added it.

Copies outside Spellwarden are your organisation's. A downloaded or pushed export is processed by the service that holds it, under your organisation's own terms with that provider and not as our processor; Claude Managed Agents, for example, stores session data in the United States. Spellwarden does not delete what was downloaded or pushed, and nothing a platform computes from an export comes back into Spellwarden.

8. The open-source scanner's usage message

The open-source Spellwarden scanner (the spellwarden command, the GitHub Action and the GitLab CI component) runs on your own computers and CI runners and, in free mode, sends your code and files nowhere. When it runs in a CI pipeline, it sends us one short usage message per run, so that we can count how many teams use it. In free mode, it sends nothing when you run it on your own computer.

The message holds exactly six fields:

• schema: the version of the message format, always 1.
• version: the scanner's version.
• rules_version: the version of its rules.
• ci: the kind of CI service: GitHub, GitLab or another.
• org: a SHA-256 hash of the account or top-level group that owns the repository on that CI service (on GitHub, the account's numeric id; on GitLab, the top-level group's path; on another CI service, the owner named in the repository's address), or nothing when the scanner cannot tell.
• a2a: yes or no, whether the run checked an A2A agent card, from the repository or from an agent you listed (since scanner 1.3.0).

This is pseudonymous data, not anonymous data: anyone who knows the account or group can compute the same hash. When the account belongs to a person, for example a personal GitHub account, the hash is personal data about that person. We cannot link it to a Spellwarden organisation or account, and we do not try. The message also carries one yes-or-no fact about the repository's content per run: whether the scanner checked an A2A agent card. No card, address or count of cards is sent. We process it on the basis of our legitimate interest in knowing how the free scanner is used (Article 6(1)(f) GDPR).

The message goes to https://dash.spellwarden.ai/api/public/ping. We keep one record per hash and per calendar week, holding the number of runs, how many of those runs checked an A2A agent card, from the repository or from an agent you listed, when it was first and last seen, the last scanner and rules versions and the kind of CI service, and we delete each record 400 days after the end of its week. The record holds no IP address. To limit abuse, our rate limiter keeps the sending IP address for one hour. Like every request to the app, the request also appears in our technical logs, kept for 90 days, which record the address it arrived from, usually one of our content delivery network's (see what personal data we collect).

To switch the message off, set the environment variable SPELLWARDEN_TELEMETRY=off or DO_NOT_TRACK=1, run the scanner with --no-telemetry, put telemetry: false in .spellwarden/config.yml, or set the telemetry input of the GitHub Action or the GitLab component to false.

9. Connected mode of the scanner

The scanner has a connected mode. When a pipeline gives it an organisation credential, it sends the results of its scan to your organisation in the Spellwarden app, so that your team sees them in the dashboard. Without an organisation credential, the scanner behaves exactly as described in the usage message section above.

In connected mode, besides the usage message described above, which is still sent from CI unless you switch it off, exactly these requests leave the runner. On your own computer with a CI key, everything below except the upload is sent. All except the first go to the Spellwarden API at dash.spellwarden.ai.

• On GitHub, the request for an identity token, which goes to GitHub's own token endpoint.
• The exchange of the code host's identity token for a session.
• The session and policy requests.
• The registry check: the path, name and identity of each skill, server, pinned command, plugin and marketplace in scope, the repository identity, commit and visibility, and the source address and pin of each plugin and marketplace.
• The upload of the scan results.

The upload holds:

• The findings, each with a masked excerpt of at most 300 characters of the lines that matched a rule, and the file paths.
• Package hashes.
• Server addresses, as hashes, except in two places: the request link for a server that is not admitted to your organisation's private registry, which carries that server's address without any user name, query or fragment; and a finding about a server's address itself (for example a server reached over plain HTTP), whose masked excerpt shows that address without the user name or password.
• The identities of the pinned commands and plugins, the inventory of what was scanned (the paths and names of skills and servers, and the result of each server probe), and the lock summary.
• The run context: the repository name, the commit, the branch, the pull request number and a link to the run.

No request carries a whole file. Notice and error messages are cut to 300 characters and masked in the same way. Where a setting refers to an environment variable such as ${VAR}, the scanner sends the name and never the value.

The code host's identity token is sent once to Spellwarden to be verified, and it is not stored. That token names the person who started the run; we use it only to verify the pipeline and keep no personal detail from it in any record or log. A CI key or CI token is sent only as the authorization header of a request to the Spellwarden API. The scanner never writes it to its output, to a log, to a comment, to a report or to the upload, and the pipeline never shows it.

For personal data about other people inside what your pipelines send, your organisation is the controller and we act as its processor under Article 28 GDPR. We process that data only on your organisation's instructions and only to provide the service.

We store a record of each repository (its name, address, visibility and default branch) while it is listed in the dashboard, the findings of its default-branch scans, and for each pull or merge request scan or key report its counts and top findings. Scan summaries of pull requests and merge requests, and reports uploaded with a CI key, are deleted after 30 days. For each repository, we keep the latest five scans of the default branch, and always the one shown as current, while the repository is listed in the dashboard, and delete older ones with their findings. A repository that was replaced on the code host and bound again starts a new history with its own five. When you remove a repository in the dashboard, the rest of its scan records are removed too. When an organisation is deleted, its records follow the rule stated under retention. Like every request to the app, these requests also appear in our technical logs (see what personal data we collect).

10. Error monitoring (planned)

We plan to add Sentry, an error-monitoring service provided by Functional Software, Inc., to the app and its backend, so that we learn about errors and can fix them. Sentry would receive technical details of an error, such as the page or request that failed, the browser and operating system, the time, and your user id or organisation id. It is not active yet. Before we turn it on, we will update this policy and the cookie policy with where Sentry stores the data, how long it keeps it and the safeguards for any transfer outside the EU.

11. Purposes of processing

• Providing the service: scans, policy packs, evals, runtime telemetry, eval exports, the private registry, API keys and the MCP server.
• Signing you in and controlling access, including single sign-on and two-factor authentication.
• Running organisations: members, roles, invitations and the audit log.
• Running the 14-day trial and billing subscriptions through Creem.
• Sending the email the service needs: sign-up confirmation and password-reset codes (sent through Amazon Cognito); invitations; a welcome email when you create an account; security notices when you change your password in Spellwarden, when your password is reset, and when we notice that two-factor sign-in was turned on or off; to owners and billing members, reminders before and when an organisation's trial ends; and notices when your role changes, when you are removed from an organisation, or when someone accepts your invitation. Organisation notices also go to accounts created through an organisation's single sign-on, at the address its identity provider gave us. You can turn off the trial reminders and the membership notices in your account settings. Security notices and codes cannot be turned off. Creem sends invoices, receipts and payment notices.
• Replying to messages you send us.
• Reviewing the rules your organisation chooses to share with us, to improve the open-source scanner's rules.
• Counting how many teams use the open-source scanner in CI (its usage message, see the scanner's usage message).
• Keeping the service secure and investigating abuse.
• Understanding which pages of spellwarden.ai are read and improving them (Google Analytics, only with your consent).
• Finding and fixing errors (Sentry, once it is active, see error monitoring).

12. Legal basis for processing

We process your data under the following legal bases (GDPR Article 6): performance of a contract (providing the service your organisation signed up for and billing it), legitimate interest (service security, abuse prevention, the audit log, counting use of the open-source scanner, reviewing and adopting the rules an organisation's manager or owner chooses to share with us to improve the open-source scanner, keeping the records of deleted organisations described under retention, and finding and fixing errors), and legal obligation (keeping financial records, and responding to lawful requests from ANSPDCP (the Romanian National Supervisory Authority for Personal Data Processing) or ANPC (the Romanian National Authority for Consumer Protection)). Usage data collected through Google Analytics is processed only with your consent (Article 6(1)(a) GDPR), and storing and reading the analytics cookies on your device also relies on your consent under Article 4(5) of Romanian Law no. 506/2004. You can withdraw your consent at any time with Cookie settings in the footer of spellwarden.ai, without affecting processing before the withdrawal.

13. Data sharing

We share personal data only with the processors needed to run the service, each under its own data processing terms, and with the export destinations your organisation chooses:

• Amazon Web Services (AWS): hosting, storage, sign-in (Amazon Cognito) and email (Amazon SES), in the eu-central-1 (Frankfurt) region. Amazon CloudFront delivers the pages and the app through its network of edge locations.
• Amazon Bedrock, an AWS service: runs the AI models of evals (see evals), on Bedrock's EU cross-region inference profiles, entirely within AWS's infrastructure. It processes your content only to produce the result. It also screens email sent to us for spam (see inbound email). AWS is the processor for these calls; Anthropic, the models' developer, does not receive your data from these Bedrock calls, and the calls are not used to train any model.
• Anthropic PBC (United States): an AI model from Anthropic, used through its API, helps us review the rules your organisation shares with us (see the terms' section on rules you share with us). Anthropic is our processor for that review and does not use the data to train its models.
• Creem (creem.io): the merchant of record for Spellwarden subscriptions. Creem is the legal seller: it takes the payment, charges the tax that applies and issues your invoice. It is an independent data controller for the payment data it collects, under its own privacy policy.
• Google Ireland Limited (Gordon House, Barrow Street, Dublin 4, Ireland): if you choose to sign in with Google, Google tells us who you are under its own terms. If you allow analytics on spellwarden.ai, Google provides Google Analytics as our processor under Google's data processing terms for Google Analytics (the Google Ads Data Processing Terms). Google Signals and ad personalisation are turned off.
• Your organisation's identity provider: if your organisation uses single sign-on, the identity provider it chose signs you in and tells us your email address and name. It is your organisation's service, not ours.
• Your organisation's export destinations: when someone in your organisation pushes an eval export (see eval exports), it goes to an account your organisation's owner added on AWS AgentCore Evaluations, Microsoft Foundry or Claude Managed Agents, under your organisation's own terms with that provider, not as our processor.
• Sentry (Functional Software, Inc.): planned, not active yet (see error monitoring).

When a scan fetches a public skill from GitHub or GitLab, or looks a server up in the public MCP registry, those services see a request from our servers, not from you, and we send them no personal data. We do not sell personal data, and we do not share it for advertising.

14. Data storage and security

Your data is stored in AWS's eu-central-1 (Frankfurt, Germany) region: encrypted at rest, encrypted in transit (TLS), with access limited to what the service needs to run. Our database keeps point-in-time backups for 35 days. Rules your organisation shares with us are also reviewed with the help of Anthropic in the United States (see international transfers). Google Analytics usage data, if you allow it, is stored by Google instead (see data sharing and international transfers).

15. International data transfers

AWS stores and processes your data in the EU for this service (eu-central-1 hosting, the EU Bedrock inference profiles). Amazon CloudFront serves each page from the edge location nearest to you. Data you fetch into an AI app you connect goes wherever that app's provider processes it, under its own terms, and an eval export your organisation pushes goes to the account and region its owner chose (see eval exports). Rules your organisation shares with us are reviewed with the help of Anthropic in the United States, under its data processing terms and the EU Standard Contractual Clauses. Creem, as an independent controller for payment data, may process billing data outside the EU under its own compliance framework (including Standard Contractual Clauses where applicable); see Creem's privacy policy. If you allow analytics, Google LLC in the USA may process Google Analytics data as a sub-processor of Google Ireland Limited. That transfer is safeguarded by the EU-US Data Privacy Framework (Google LLC is certified) and by Standard Contractual Clauses.

16. Retention

• Your account: kept until you delete it. When a person deletes their account, their profile and sign-in methods are erased and their past actions in organisations remain as anonymous entries. Deleting your account also removes your connected apps.
• Sign-ups never confirmed: deleted after 24 hours.
• Deleted organisations: when an organisation is deleted, Spellwarden keeps its records (settings, member list, invitations including the invited email addresses, audit log and billing references), never shows them to anyone and never gives its URL name to another organisation. For one day after the deletion, an owner can restore the organisation. After that day it stays closed. Invited email addresses are removed with their invitation (below); addresses in the audit log and in the record of removed members stay. You can ask us to erase personal data about you held inside a deleted organisation by emailing hello [at] spellwarden [dot] ai.
• Invitations: an invitation can be accepted for 30 days. Its record, with the invited email address, is deleted 30 days after that.
• Audit log and billing references: kept for as long as the organisation's records are kept.
• Content your organisation submits: kept while the organisation exists, unless the organisation deletes it. Uploaded files that were never scanned are deleted after one day. The masked text sent to an eval run is kept for 400 days, the AI models' answers for 30 days, and policy pack exports for 30 days.
• Shared rules: a share under review is kept until we decide or you withdraw it, or until your organisation is deleted. A rule we do not adopt is reduced to our decision and its area and severity. For a rule we adopt we keep the rule and our decision, also after your organisation is deleted, without the note or where the skill was published.
• Connected apps: a connection ends when you revoke it, when it reaches 30 days, or when its refresh token expires unused.
• Inbound email: the full copy in our AWS storage is deleted after 90 days. Our conversation with you in our mailbox is kept for 365 days, so we can find its history.
• Technical logs: API Gateway and app logs, and the website logs of spellwarden.ai, are kept for 90 days. The rate limiter keeps an IP address for about a day.
• Backups: database backups roll over after 35 days.
• Scanner usage records: deleted 400 days after the end of the week they count.
• Connected mode scans: scan summaries of pull requests and merge requests, and reports uploaded with a CI key, are deleted after 30 days. For each repository, the latest five default-branch scans are kept while the repository is listed. See connected mode of the scanner.
• Runtime telemetry: session details 30 days, daily totals (with the names, addresses, URLs and skill sources of what was used) 400 days. Stored prompts, answers and tool content 30 days, and deleted within two hours when content storage is turned off; session eval transcripts are deleted when the run ends, and the judge's explanations after 30 days. See runtime telemetry.
• Eval exports: 30 days. An export of recorded sessions is content: 30 days, and deleted within two hours when content storage is turned off. Copies downloaded or pushed are your organisation's to delete. See eval exports.
• Google Analytics event data: 14 months, then deleted automatically by Google Analytics.

17. How results are produced

Scans are automated checks of the skill packages and MCP servers your organisation submits. Evals send the chosen skills' instructions, or a server's tool descriptions, with Spellwarden's test tasks to AI models running in EU regions only and grade the answers (see the section on evals above); their results are informational. No decision with legal or similarly significant effect on you is made by these models; your organisation decides what to do with a result. We do not read your organisation's content except for support, security or legal reasons, and to review the rules your organisation shares with us.

18. Your rights

Under the GDPR, you have the right to: access the personal data we hold about you; request correction of inaccurate data; request erasure (the right to be forgotten); request restriction of processing; data portability; and object to processing based on legitimate interest. You can withdraw a consent at any time. To exercise any of these, email hello [at] spellwarden [dot] ai. You also have the right to lodge a complaint with ANSPDCP (the Romanian National Supervisory Authority for Personal Data Processing) (see complaints). Where your organisation is the controller of the data, we pass your request to it and help it answer.

19. Minors

Spellwarden is for businesses and adults. It is not directed at, and we do not knowingly collect data from, individuals under 18.

20. Data Protection Officer (DPO)

NETBEARS TEAM SRL has not appointed a statutory Data Protection Officer, as our processing does not meet the GDPR's mandatory-DPO thresholds. Data protection questions can be sent to hello [at] spellwarden [dot] ai.

21. Complaints

If you believe we have not handled your personal data properly, please contact us first at hello [at] spellwarden [dot] ai. You may also lodge a complaint with ANSPDCP (the Romanian National Supervisory Authority for Personal Data Processing): https://www.dataprotection.ro/.

22. Changes to this policy

We may update this policy as the service changes, and we will update it before Sentry error monitoring goes live. The date at the top of this page is when it was last updated; material changes will also be noted on this page.

Changes on 2 November 2026: a manager or owner can now share one of their organisation's rules with us. This policy adds the review of shared rules as a purpose (see purposes), Anthropic as our processor for that review and the transfer to the United States it involves (see data sharing and international transfers), and how long shared rules are kept (see retention).

23. Contact

Company

NETBEARS TEAM SRL

Trade Register

J2017000687229

EUID

ROONRC.J2017000687229

VAT ID

RO37261366

Address

Corneliu Baba 8, Iași, Romania

Phone

+40 742 121 246

Email

hello [at] spellwarden [dot] ai