Project-data handling

What is stored

The TMS stores uploaded documents and working files in object storage. Application records include source and translated segments, revisions, projects, workflow information, translation memories, and terminology. Depending on the features you use, processing also creates exports, cached translations, conversations, analysis and review reports, and evaluation records containing source and candidate text.

Those derived records matter: project content can exist in more places than the original uploaded file. Account authentication is provided through Supabase Auth; the main project database is application-owned. The marketing website’s hosting and enquiry-retention arrangements do not describe the TMS’s project-data lifecycle.

For the declared primary hosting architecture and access controls, see Security overview.

AI and translation services

The current implementation supports Google Gemini and OpenAI for AI functions, including AI-generated machine translation, and DeepL for machine translation. Gemini and DeepL can use your organization’s own API key or, where the platform provides one, a platform key; OpenAI is used only with your organization’s own key. Support for a provider does not mean it is enabled for your organization. We need to confirm which services, accounts, and features are enabled for your organization before presenting a definitive processor list.

When you run an enabled AI or translation feature, the selected provider can receive the text and context required for that task. This may include source text, candidate translations, terminology, translation-memory examples, surrounding text, reference material, instructions, and conversation or tool messages. It is not accurate to describe every request as anonymous metadata or a single isolated segment.

Organization-supplied credentials and platform-provided credentials can lead to different provider accounts. A customer-supplied API key does not, by itself, establish a particular processing region, no-training protection, or zero retention. Applicable terms depend on the actual service, account, and enabled features; we do not make a blanket zero-retention promise for external AI providers.

Diagnostics and transcripts

Application diagnostics: Sentry integration includes sanitization and controls to mask text and inputs, block sensitive project-content areas and media in replay, and avoid recording network bodies. Session replay is controlled by organization admission and eligible routes. These controls reduce collection; they do not establish that all monitoring data is anonymous. Your organization’s diagnostic configuration should be confirmed with us. Browser diagnostics and admitted session replays can include the signed-in user’s application identifier and the organization’s identifier and name, so support can associate an event with the correct account. These identifiers are not anonymous, even when project text is masked. Disabling Replay does not by itself disable error diagnostics.

Optional customer-connected Langfuse: Your organization can connect its own self-hosted instance or Langfuse Cloud project. When enabled, TilBridge emits integration events that can contain AI prompts, responses, tool messages, and selected evaluation scores to that destination. TilBridge does not host or maintain a separate Langfuse-style trace store on your behalf; trace storage and presentation take place in your connected Langfuse instance. Its hosting, access, retention, and deletion settings are separate from TilBridge’s. This does not mean that TilBridge stores no project content: application records and core evaluation facts described above are retained independently of this integration.

Support: Information you deliberately include in a support message is another disclosure. Start with a brief description rather than attaching private project content or credentials.

End of evaluation and deletion

When an evaluation ends, access can be restricted; that should not be understood as automatic deletion of all customer data. Ordinary job deletion deliberately preserves some reusable assets and historical evidence, including translation-memory or terminology entries and evaluation snapshots. Backups and external-provider copies have separate lifecycles.

Before an evaluation with your own material, agree with us on the return or export method, the handling when the evaluation ends, the deletion scope and timetable, and any bounded backup or legally required retention. A complete request needs to address relevant derived records and integrations as well as uploaded documents. Do not assume a download window after access ends unless it has been arranged.

We are not publishing a universal deletion deadline here while those arrangements require confirmation. Contact privacy@tilbridge.kz to discuss retention, return, or deletion. Please identify your organization without including the project data itself in the initial email.

Your arrangements

Tell us whether you process on behalf of another organization and whether your material contains personal or sensitive information. Before uploading confidential or personal data, confirm the enabled providers, permitted processing locations, and applicable agreements with us.

See DPA enquiries for the information needed to discuss a processing agreement. The website privacy notice separately describes this marketing site’s enquiries, preferences, and analytics.