coplug
    Trust Centre

    Legal document

    Data and Model Use Notice

    How Coplug uses customer content, memory, derived representations and anonymous data.

    Version 1.0 · Effective 01.10.2026

    Clause 11 of our General Terms and Conditions sets out how we use your content. This notice explains it in more detail.

    It forms part of your contract with us. Clause 22.3 of the Terms limits what we can change here: a change to this notice will only ever be within the purposes already stated in the Terms. A new purpose needs a change to the Terms themselves.

    If anything here conflicts with the Terms, the Terms apply.

    The four things that happen, and why we keep them apart People ask one question about an AI product: what happens to what I put in. The honest answer has four parts, and running them together is how most policies end up misleading.

    What it is               Contains your           Can you opt out?
    content?
    Memory                  A record of your         Yes                     It is the product.
    work, kept so Coplug                             Delete the content
    can show it back to                              and the memory goes
    you                                              with it
    Training                Using your prompts       Yes, during training.   Yes. Enterprise by
    and questions to         Never your              agreement, everyone
    improve Coplug's         uploaded files or       else by objection
    own models               datasets
    Derived                 Embeddings and           Transformed, not        Enterprise by
    representations         indexes, with direct     original, and not       agreement
    identifiers stripped     linked back to you
    Anonymous Data          De-identified records    No                      No, and it does not
    of how the product is                            contain your content
    used

    1. Memory

    Coplug keeps a record of your work: what you looked at, what has changed in data you engaged with, and the reasoning behind outputs you created.

    We also use these records to understand how Coplug is used and to improve it, as described in the rest of this notice.

    Your memory is stored against your account. No other account can read it. Delete a conversation or close your account and the memory goes with it.

    2. Training our models

    What we train on. Your prompts and questions, used to train, fine-tune, evaluate and improve Coplug's own models, prompts, retrieval indexes and configurations.

    We do not use files or datasets you upload to train our models. There is no pipeline that does this.

    What we are actually doing today. We are not running model training on customer content in production. The work in development is on how Coplug reaches a good answer, learning from the sequence of steps that led to a successful result rather than from the content of the question. That carries nothing that identifies you or your organisation.

    The paragraphs above describe what the Terms permit. If we start doing more than we describe here, we update this notice first. Clause 22.3 limits any change to the purposes already set out in the Terms.

    What we do not do. We do not use your content to train the foundation models of any third party. The providers whose models we build on are contractually prevented from using what we send them to train their own.

    Our legal basis. Where your content is personal data, we rely on our legitimate interests under Article 6(1)(f) UK GDPR, supported by a documented Legitimate Interests Assessment.

    Your right to object. You can object at any time under Article 21 UK GDPR by emailing privacy@coplug.ai. If you object, we stop using your content for training. This right does not depend on your plan and it is not a paid feature. It applies to everyone, including people on a free plan.

    Enterprise. Enterprise customers can agree in their Order which categories of their content are excluded. The categories are set out in the Plan and Feature Terms.

    What training does not do. It does not create a lookup of your content that another customer can query, and it does not make your work searchable by anyone else.

    3. Derived representations

    Content you submit, and records of how you use the product, may be turned into derived representations such as embeddings and indexes. Direct identifiers are removed or obscured before those are used for anything beyond running the service for you.

    We use them to improve retrieval, memory, personalisation and performance. Where derived representations feed our own development work, they do so in that de-identified form.

    Nothing links a fragment back to you. A derived representation carries no route back to the account, the organisation or the source file it came from. That separation, rather than the mathematics of the representation itself, is what protects you.

    Embeddings are not stored text. They are numerical representations, and nobody can open one and read it. They are also not treated by us as anonymous. We hold them inside the same access controls as the content they came from, and they are never exposed outside our infrastructure.

    4. Anonymous Data, and the aggregation gate

    What it is. De-identified records of how the service is used. That an action was taken, and the sequence of steps within a workflow.

    It does not include your content. It is not your prompts with the names taken out. It is the shape of the interaction rather than its substance.

    We collect it on every plan including Enterprise, and we own it.

    The aggregation gate. We may use Anonymous Data to develop insight about patterns across customers. We will not present any insight drawn from more than one customer's activity unless it has first passed our anonymisation and irreversibility checks, and has been assessed as not reasonably capable of identifying any customer, organisation or individual.

    We apply a minimum threshold of distinct customers before any cross-customer insight is shown, and a higher one where the population being described is small enough that a low count would point at a single organisation.

    What we do not do. We do not sell your content or Anonymous Data. We do not disclose Anonymous Data to anyone except in aggregated form that has passed the gate above.

    5. Where Coplug runs

    Our core infrastructure is hosted in the United Kingdom.

    Prompts are processed by AI model providers under contracts that prevent them using that content to train their own models. A current list of our providers is available on request during procurement.

    We keep this list outside the Terms on purpose. Naming a provider in a document you have to accept again means a supply chain change forces re-papering. Here it can be updated with a version bump, and clause 22.3 stops that flexibility being used to widen the purpose.

    6. What deletion reaches

    Clause 11.7 of the Terms sets out how deletion works.

    Deleting your content does not reverse improvements already made to our models. Once your content has contributed to a training run, the resulting model cannot be unpicked back into the individual contributions that shaped it.

    The safeguards above are applied before content reaches training. Anonymous Data derived from your activity also remains, and it contains none of your content.

    If you object to training, we stop using your content going forward. We cannot retrospectively remove its contribution to a model already trained.

    7. How to exercise your controls

    What you want How

    Object to your content being used for training privacy@coplug.ai

    Exclude categories of content as an              Agree them in your Order
    Enterprise customer

    Delete a conversation or a file In the product, at any time

    Delete everything Close your account

    Export your outputs                              Use the export function before closing your
    account

    Ask what we hold about you privacy@coplug.ai