Data handling and deletion
What Vidext stores, including chat conversations and your organization's course files, which data is filtered, and the practical boundaries of retention and deletion.
Use this page to see what Vidext keeps when your organization uses it, who can read it, and what deletion does and does not remove. It covers data stored and controlled by Vidext; for data sent to models, media services, and research providers, see AI providers and data flow.
Two terms come up throughout. The chat is where admins and members build courses by telling Vidext's agents what to teach. Vidext's agents are the AI agents that research your material, plan the course, write each module (a short interactive video lesson) and build it.
Data stored by the product
| Category | What Vidext may store |
|---|---|
| Account and organization | Email, profile and preferences, organization membership, roles, team assignments, and security-session records. |
| Sources | Uploaded originals and video links, pages read from your company website, extracted text, transcripts, descriptions of what videos show, page images, source metadata, and processing status. |
| Chat conversations | Your messages, the agents' replies and a record of the actions they took, attachments, and who was invited to each conversation. |
| Course files | Course plans, module scripts and their earlier versions, review results, the organization's knowledge and brand files, and what the agents remember for each person they work with: their lasting preferences about how they build courses and the facts about the organization they confirmed. They are kept in a private store for each organization that records a history of every saved change (a version-controlled repository). |
| Generated content | Built and live module versions, the record of each course publication, branding, images, narration, audio, video, and their provider identifiers. |
| AI tutor conversations | Learners' messages, the tutor's answers, and the feedback needed to keep the conversation's history. |
| Learning records | Course enrollments, module completions, sessions, progress, visited scenes, time spent, quiz selections, correctness, and attempts. |
| Operational records | Usage, cost, status, errors, cache state, request and trace identifiers, and activity or security events. |
Content and records are scoped to an organization and checked against its membership and role model. See Access control.
Chat conversations and course files
Only admins and members can open the chat. See Working in the chat for how it works.
- Who can read a conversation. The person who started it and the admins and members brought onto it: by the owner from the Share panel, or by anyone already on the conversation who tags them in a message. A person added sees the whole conversation, including what was said before they joined. Organization admins cannot open a conversation they were not added to. See Work together.
- Tagged messages are copied into notifications. When a message tags somebody, its text is copied into their notification: the notice in their Inbox keeps up to 500 characters of it, and, if the person has turned on email in their notification preferences, the email sent right after the tag carries the message. Inbox notices are removed when the conversation is deleted, or when the person is removed from it or leaves; an email already sent keeps the message in the recipient's mailbox even after the conversation is deleted, and Vidext cannot withdraw it.
- Attachments become sources. A file attached in a conversation becomes a document of the organization, which every conversation can draw on.
- Each conversation works on its own copy. While the agents work, they change a private working copy of the organization's course files that belongs to that conversation. Saving the conversation's work writes it back to the course files, recorded under the conversation's owner and anyone else who worked in it.
- What the agents remember is not sealed to one person. Each person's memory is kept in the organization's course files. The agents load only your own memory into your conversations, but they can open a colleague's if asked to in an admin's or member's conversation. See Working in the chat.
- How long conversations are kept. History is kept until the conversation's owner deletes it; archiving keeps it, and a person's Main thread cannot be deleted. Deleting a conversation removes its messages and its list of invited people; work already saved to the course files stays. A conversation's working copy is cleared after 14 days without activity, and anything in it that was not yet saved is lost.
- When someone leaves. Removing a person from the organization, or changing them to learner, ends their access to every conversation, and the people they invited into their own conversations lose access too. Removing access does not delete the conversations.
Storage boundaries
- Originals and derived data are separate. Uploaded source files are kept separately from extracted text, transcripts, images, and other derived files used during generation.
- Live modules are stable versions. The version delivered to learners is stored independently from later edits so learning activity can be associated with the content version that was served.
- Generated media has durable references. Vidext stores generated images, narration, audio, and video when they are part of workspace content, together with the metadata needed to deliver or regenerate them.
- External processing is separate. A provider may temporarily process the data needed for a feature even when Vidext stores the resulting content in its own systems. See AI providers and data flow.
Controls that reduce unnecessary data
- Activity metadata is filtered before storage. Known sensitive keys such as email, name, password, prompt, token, content, and authorization are removed from activity metadata, and the remaining object is bounded by depth and size. This is a defensive filter, not a guarantee that arbitrary metadata can never contain personal data.
- Connector credentials are encrypted. Stored authorization values for connected services are encrypted separately from ordinary connector metadata. Disconnecting a connector disables future access through that connection.
- Security logs and workspace activity are separate. Authentication and session events can include technical context such as IP address and user agent; ordinary organization activity uses a separate event history.
- Product analytics is content-restrained. The application associates analytics with user and organization identifiers, but session replay is disabled and captured text and attributes are masked by default.
Sessions and connected services
Expired sessions and verification records are removed by a scheduled maintenance process. Users can review and revoke active sessions in security settings; contact infosec-external@vidext.io if account or device access needs additional review.
The Sessions page is open to admins and members and is not listed in the settings menu today; the link above opens it.
Each person manages their own connected services in Settings → Connectors; a connection belongs to the person who made it, and nobody else can disconnect it. Disconnecting stops future access through Vidext; the external provider may also offer its own authorization-revocation controls.
Deletion boundaries
Deleting a product item can involve both database records and stored files. Vidext runs cleanup for supported document and media operations, but a visible deletion should not be interpreted as an assurance that every backup, provider-side temporary copy, derived file, or legally required record disappears at the same instant.
Likewise, removing a person's access and completing a personal-data deletion request are different operations. Historical, security, billing, or contractual records may require separate treatment or retention.
To request access, correction, export, restriction, or deletion of personal data, contact datos@vidext.io. Workspace return and deletion at the end of a customer agreement are handled under the applicable contract and Data Processing Agreement.
Model-training boundary
Vidext uses workspace knowledge to produce content and answers for that organization. Vidext does not use Customer Content to train its own AI models. Provider use of Customer Content is governed by the applicable agreement and Data Processing Agreement; provider-specific data flows are listed in AI providers and data flow.
Formal terms
- Privacy Policy — personal-data purposes, rights, transfers, and legal retention terms.
- Terms of Service and Data Processing Agreement — customer instructions, processor duties, and termination commitments.
This page describes current product behavior and does not replace those documents.
Last updated on