Security
We handle your Confluence and Notion content with care, because we know how sensitive internal documentation can be. Here’s exactly what we do and what we don’t yet do, so you can make an informed call. Region-specific points are at the end, under For EU residents and For non-EU residents.
What we do
- Encryption. Confluence and Notion content is encrypted in transit (TLS 1.2 or higher) and, in our AWS account, at rest. The database and its backups are encrypted with AES-256 under a KMS key we manage; storage buckets, secrets and the short-lived sign-in tables use AWS’s own managed encryption; log groups use CloudWatch’s default encryption, with no key of ours. The text we store in our database is also encrypted column by column, so someone who got into the database itself would see ciphertext. Search doesn’t read your text. We hold the key, and the running service can decrypt with it. The numeric index, keyed hashes used for search and matching, source and team names, and your site address aren’t encrypted this way. Vendors that hold content use their own encryption.
- Data minimization. We don’t keep a copy of your wiki: pages are read live from your source on each scan. What we keep is text derived from your pages, which can repeat their wording, and a numeric index of them; questions and answers on a finding, and comments you ask us to post; and, when a fix is previewed, the passage it would replace and its proposed text, so it can be applied and reverted. For a duplicate-pages merge, that is the whole text of each page merged away. We write nothing back without your explicit approval, and content fixes revert in one click.
- Tenant isolation. Customer requests run as a database role that cannot bypass PostgreSQL row-level security, pinned to their team. Every table that holds team data has that protection. The one table without it lists the teams themselves, and our code filters it. A few named system jobs, such as sign-in lookup, token rotation and scan scheduling, read specific tables across teams, and the operator’s admin credential bypasses row-level security, though it sees encrypted columns only as ciphertext.
- Minimum-necessary scopes. We request more than “read and write pages”, so here is the whole shape of it: read the spaces you scope a report to, write to pages you choose to fix, post a comment when you ask a page owner to decide and delete it if you undo that, read page metadata, and resolve directory information – users, groups, and the report owner’s email address – to work out who owns a page and where to send a scan email. A small app-storage scope lets the WikiFix agent in Rovo keep one pending summary per person in Atlassian’s Forge storage. For WikiFix for Confluence, the full list, with our rationale against each one, is public in the Forge manifest on our Marketplace listing – verify it rather than take our word for it. At app.wikifix.ai, Atlassian and Notion show the person connecting a source what WikiFix asks for before they agree, and scans read only what that person can see.
- Encrypted credentials. Your Atlassian tokens and any AI key you bring are envelope-encrypted with AWS KMS (a KMS-managed data key plus local AES-256-GCM); the key rotates automatically. Our API never returns them, plaintext travels only over TLS to the vendor it belongs to, and we don’t log them by design.
- No training on your data. Default inference runs on AWS Bedrock. Per AWS Bedrock’s terms, your prompts and completions are never used to train base models, and no model vendor has access to them – AWS is the inference subprocessor. If you bring your own Anthropic key, those calls go to Anthropic directly under your own agreement – see “Where we process your data.”
- We don’t read or reuse your content. We don’t tune or improve our AI prompts on it. A person at WikiFix reads it only when you ask us to in writing or give explicit written permission, for the problem you raised.
- Access to your content. One person operates WikiFix and administers production, under a duty of confidentiality. We keep your request or permission. We don’t keep a technical log of each read, in our tools or in a vendor’s own console, such as Langfuse’s.
- The Rovo agent. In WikiFix for Confluence, when a person uses the WikiFix agent in Atlassian’s Rovo, Atlassian’s Rovo AI receives finding text, page titles and the passages behind them, the text of any page the agent reads for them and, when they settle a finding, the before and after text. Your page bodies are already at Atlassian, and the same happens on every Data handling level, Strict included. To settle a finding the agent keeps one summary per person in Atlassian’s Forge storage until they settle it or prepare another; we set no expiry on it while WikiFix is installed, and our deletion job doesn’t clear it. Atlassian deletes app storage after an uninstall, on its own retention schedule.
- Subprocessors. Each is governed by its standard data-processing terms, and every one that offers an EU region is configured to it, with two exceptions. TypeSafe has no EU region and processes in the US: installs that require EU residency never reach it and run its two checks on eu/jev in Germany instead. For the rest, TypeSafe keeps none of what it receives and uses none of it for anything else, including training AI models – see “Where we process your data”. Atlassian holds contact-form submissions from this website without a residency pin. Product analytics (PostHog) receives events with no page content: an allow-list per event on our servers, a block-list of content and external-id fields in the browser. Comments you choose to write in feedback go there too, and your reply-to address if you leave it ticked. Events also carry page addresses, browser and approximate country. Session replay masks text and inputs, leaves out elements that hold your text in an attribute, and records layout and clicks. (Analytics for this marketing website collects different things – see the Privacy Policy.) Error tracking (Sentry) receives stack traces, error codes and page addresses, with no page content, names or emails, and performance traces: the timings of requests and page loads, with addresses but no query strings. Langfuse receives, on the Standard level, the full text we send to the AI model and the model’s response, which include the page content under review, so we pin it to an EU-resident instance and delete that text about 30 days after it’s recorded. The full list, with each vendor’s purpose and location, is on our Sub-processors page; changes are published in advance to a feed you can follow.
Data handling
Your team chooses a level under Admin → Data handling. During onboarding, the first admin to answer is asked about sensitive data and EU residency, and later admins aren’t asked again. It starts on Standard. We keep what WikiFix needs to work on every level: the database, account and billing records, and server logs. The level decides everything else:
| Standard | Restricted | Strict | |
|---|---|---|---|
| AI requests and responses, at Langfuse | Full text, about 30 days | Cost and timing with our internal ids, and no text | Not recorded |
| Two analysis checks, at TypeSafe in the US | Only without EU residency | Only without EU residency | Only without EU residency |
| The same two checks, at eu/jev in Germany | Only with EU residency | Only with EU residency | Only with EU residency |
| Error reports and traces, at Sentry | Sent, with no page content | Sent, with no page content, site address, email, name or Atlassian or Notion account id | Not sent |
| Usage events, at PostHog | Sent, with no page content | Sent, with no page content, site address, email, name or Atlassian or Notion account id | Not sent |
| Session replay, at PostHog | Recorded, with text and inputs masked, and elements that hold your text in an attribute left out | Recorded, with text and inputs masked, and elements that hold your text in an attribute left out | Not recorded |
| Feedback a person writes to us, at PostHog | Sent if they send it, with their email if they leave “Reply to me” ticked | Sent if they send it, without their email | Sent if they send it, without their email |
| Report and space names in scan emails, at Brevo | Sent | Not sent: counts and links only | Not sent: counts and links only |
If your content holds special-category or criminal-offence data, such as health data, you must use Restricted or Strict. We don’t look at your content, so we can’t tell. If you use Standard with such data anyway, that is your decision and your responsibility.
On Restricted and Strict, we can’t look into past AI results if you report a problem. Moving to either deletes your team’s records at Langfuse, usually within about a day and longer for a team with many records. The level doesn’t change what a scan costs. Admin → Change history shows each change of your configuration, permissions and consent, and each share link created, revoked or opened. It doesn’t list reads of your content.
Where we process your data
Customer data is stored in AWS Frankfurt (eu-central-1), and compute runs there too. Default AI inference runs on AWS Bedrock through EU inference profiles, which can serve a request from any EU region; our analysis models refuse cross-continent profiles.
With EU residency required on the Residency tab in WikiFix Admin (see Data residency), every AI step stays in the EU, and a scan that cannot meet that stops rather than quietly processing your content elsewhere. Two analysis checks help decide whether a finding stands. With EU residency, they run on different AI models, which reside in the EU: eu/jev, operated by Bevelites GmbH on its own server in Germany. Every other AI step stays on AWS Bedrock in the EU. Without it, on every Data handling level, they run on TypeSafe, whose endpoint processes in the United States. Either way, a request receives excerpts of the pages it checks, which can come from several pages: each passage with its page’s title and headings and the section around it, up to 2,500 characters of section. On a short page that can be the whole page; it never receives the whole space. eu/jev doesn’t store the excerpts it receives, in its database or its logs, and doesn’t use them to train AI models. We use TypeSafe with zero data retention: it doesn’t keep the excerpts it receives or use them for anything else, including training AI models. Today we run one region, and EU is the only residency we can serve.
Sentry and PostHog store their data in the EU, but both are US companies and can process some of it in the US, under the EU-US Data Privacy Framework and the EU Standard Contractual Clauses. We send neither of them page content. Feedback a person writes reaches PostHog as they wrote it.
The other path that leaves this setup is the one you choose: the optional bring-your-own-key path, available on a paid Advanced plan in WikiFix for Confluence and on the Business plan at app.wikifix.ai. If you configure your own Anthropic API key, the AI calls made on it, the checks above included, go to Anthropic’s API directly, whatever residency you’ve set, outside the EU-region pin and under your own agreement with Anthropic. Indexing stays on AWS Bedrock, and those calls are still recorded to Langfuse as your Data handling level sets.
Government access & legal reach
WikiFix is operated by Totem Dev, incorporated in Moldova. Where authorities can compel data, they reach it through the providers that hold it and through legal-assistance channels; our place of incorporation does not put it beyond reach.
- Your data sits with AWS, a US provider. AWS can be compelled under the US CLOUD Act wherever it stores data, including Frankfurt. AWS’s published commitments include challenging government data-access requests and notifying customers when legally permitted, but the legal reach exists.
- Authorities can ask Moldova for assistance. There’s no bilateral US–Moldova treaty, but both are parties to multilateral mutual-legal-assistance conventions – the Council of Europe Budapest Convention on Cybercrime, the UN Convention against Transnational Organized Crime, and the UN Convention against Corruption.
What actually protects your data is EU-region processing, data minimization, and encryption.
Data retention and deletion
We delete your data about 30 days after you uninstall WikiFix for Confluence or delete your team at app.wikifix.ai. This is automatic – a daily job starts the deletion on its first run past the 30-day mark, whether or not you ask, once every install on the team is uninstalled. Uninstalling stops our scans of your Confluence site at once and deletes the app token we cached for it. Notion workspaces and other Confluence sites that someone added by hand stay connected for those 30 days, with their scheduled scans waiting, and their sign-in tokens are deleted when the 30 days are up. Disconnecting a source deletes the sign-in tokens we hold for it, and nothing already stored from it, which stays until the team is deleted. Deleting a team at app.wikifix.ai deletes its tokens at once. When someone is removed from a team, we stop using their sign-ins. We ask Notion to revoke the Notion tokens whenever we delete or retire them. Atlassian gives apps no way to revoke a token, so a person’s WikiFix grant stays listed under Connected apps in their Atlassian account until they remove it. The 30 days exist so a re-install picks up where you left off. Want it gone sooner? Email security@wikifix.ai. Database backups have a 7-day retention window.
That job covers our database and sign-in records, apart from your change history and the ids that tie it together (ours, Atlassian site, account and group ids, and Notion workspace ids), which we keep, each entry for three years from its date, even after your team is deleted, as a record of what applied when. Email security@wikifix.ai and we’ll delete it sooner. Deleting a team doesn’t reach what sits elsewhere, which runs on its own clock: text in Langfuse is deleted about 30 days after it’s recorded, or usually within about a day of moving to Restricted or Strict, and longer for a team with many records; server logs, which include IP addresses, internal ids and our vendors’ reference numbers, expire after 90 days; analytics events go after 3 months, at PostHog’s next monthly cleanup; error data after 30 days, and session replays after 30; database performance records after 15 months; the terms of Brevo, which delivers our emails, set no fixed retention period. TypeSafe keeps none of the excerpts it receives and uses none of them for anything else, including training AI models. eu/jev, which runs those checks with EU residency, doesn’t store the excerpts it receives. What your browser caches is on your device and stays until you sign out or clear it.
Data Processing Agreement
We’re your processor, so our Data Processing Agreement, with the EU Standard Contractual Clauses, applies automatically when you accept our Terms – there’s nothing to sign. If your procurement process needs a countersigned copy, email security@wikifix.ai. It describes all three Data handling levels, and choosing one is your instruction to us.
What we don’t have yet
SOC 2. We don’t hold a SOC 2 attestation yet, so if your procurement requires one today, we’re not the right fit yet. We plan to obtain one, and this page will say when we have. We answer custom security questionnaires within 3 business days.
EU-sovereign AI by default. Default inference runs via AWS Bedrock in the EU, on models AWS hosts and serves itself, so no prompt is sent to the model vendors. Those vendors do not see your prompts or completions; AWS is the inference sub-processor. Two caveats. The first is AWS itself – the inference stays in the EU, but the company serving it is US-incorporated. The second applies only to installs without EU residency: the two analysis checks described under “Where we process your data” run on TypeSafe in the US, a third-party vendor that sees the passages it checks, though it doesn’t keep them or use them for anything else, including training AI models. Requiring EU residency removes the second caveat, not the first: those checks then run on eu/jev, operated by Bevelites GmbH in Germany, which sees the same passages and doesn’t store them.
If you require an EU-incorporated model provider, WikiFix can also run on Mistral (French, EU-incorporated, on its EU endpoint in France). We switch that for a whole residency tier rather than one team, so we can’t turn it on just for you today – tell us at security@wikifix.ai if it’s a deal-driver.
For EU residents
If you are in the EU or EEA, the GDPR applies and your organization is the data controller, with WikiFix (Totem Dev) acting as processor. Your data is stored in the EU (AWS Frankfurt), and with EU residency required for your install all AI processing stays in the EU too, unless you’ve also added your own Anthropic key; without that requirement, two analysis checks run in the US, as described under “Where we process your data”. Our GDPR-aligned Data Processing Agreement applies automatically when you accept our Terms, with nothing to sign. You may lodge a complaint with your national supervisory authority. Totem Dev’s home jurisdiction, Moldova, is a party to the Council of Europe’s Convention 108 on data protection and has ratified the protocol that modernises it (Convention 108+), which is not yet in force. We administer WikiFix from Moldova, which the GDPR counts as a transfer even though nothing is stored there; the Privacy Policy says what covers it. Data-subject rights and how to exercise them are in our Privacy Policy.
For non-EU residents
Everything under What we do and Where we process your data applies to you unchanged: your data is hosted in the EU regardless of where you are, and we extend the same controls and protections to all customers. US residency is not something we can serve today – our infrastructure is pinned to Frankfurt, so a US requirement is refused honestly rather than served from the EU and called US. You can still record the requirement; scans stop until we can meet it. If full-stack, CLOUD-Act-free hosting on an EU-incorporated provider is a hard requirement, we’re not the right fit today; tell us at security@wikifix.ai if it’s a deal-driver. US residents: see the California/CPRA note in our Privacy Policy.
Reporting security issues
security@wikifix.ai. We acknowledge within 24 business hours. We credit researchers in our release notes and respond seriously.
Contact
Our DPA needs no signature – accepting the Terms accepts it. For a countersigned copy, custom questionnaires, or specific security questions: security@wikifix.ai.