Pimenko AI — security and data for your Moodle

Some data is processed outside Moodle

Prompts, attachments and contextual information needed to answer a request may be sent to the associated processing service. Identity data is pseudonymised before this transfer. Moodle remains the starting point, return point and authority for roles, permissions and pedagogical decisions.

⌁

Infrastructure

The Pimenko AI service is hosted on Scaleway infrastructure. Scaleway operates infrastructure in the European Union, including France, the Netherlands, Poland and Italy.

✦

Certification

Scaleway publishes its certifications and security measures in its security and compliance centre, including ISO/IEC 27001.

↻

Models

Processing uses Open Weight models and a service layer developed by Pimenko.

Where is your data hosted?

A clear chain, from Moodle to the answer.

User data, content and learning records remain in your Moodle instance under your institution's rules. To answer a request, the service receives only what is needed: the prompt, useful context and, where the mode provides for it, an attachment or selected resource.

Moodleuser, course, activity, resources

→

Pseudonymisationidentity replaced before external processing

→

Open Weight AI modeland services developed by Pimenko

→

Answerreturned to Moodle, with sources where available

What remains in your Moodle instance

User data, roles and permissions
Course content and activities
Conversations retained by Moodle where the mode provides for it
Records needed for privacy, deletion and auditing

What is sent for AI processing

The prompt and context needed to answer
Resources or files needed for the requested use
Pseudonymised identifiers, never the raw Moodle identity
The mode and activity type needed for processing

Pseudonymisation

How does pseudonymisation work?

Before any transfer to an external model, directly identifiable personal data is replaced by pseudonymised identifiers. Names, email addresses and raw Moodle identifiers are not sent to the model within the Pimenko AI framework.

If a learner asks for help with an exercise, the service may receive the question content, course context and current activity, but not the learner's raw Moodle identity. Pseudonymisation does not make educational content anonymous: a message or file can still contain sensitive information.

Pseudonymisation is applied by the processing chain. The administrator controls modes, permissions and usage scope; the pseudonymised fields and configuration options should be documented with the technical team before production deployment.

Protect identity, not hide content.

Transparent data flows let institutions choose uses appropriate to their context.

Model training

Is your data used to train AI models?

No. Within Pimenko AI's contractual scope, received data is used to process the request and operate the assistant. It is not intended to train or fine-tune a third-party model.

This clause must be confirmed in the service terms and applicable DPA. A change of provider, model or retention policy must be documented and communicated to the institution.

A commitment to verify in contractual documents

Model, hosting, retention and subprocessors choices must remain verifiable by the institution and its DPO.

Governance

GDPR and AI Act compliance.

RGPD

The institution is generally the data controller: it defines the purposes and means. Pimenko acts as a processor within the applicable contractual framework. A DPA can document roles, data flows and security measures.

No automated decision-making

Pimenko AI proposes an answer, activity or prepared action. The system does not make automated decisions affecting learner rights: no admissions, grading, automatic pathway allocation or silent course changes.

AI Act

The system is designed as non-decision-making assistance, not as an assessment or training-access tool. Its risk category must nevertheless be legally confirmed according to the actual purpose, users and selected configuration.

Retention and deletion

Configurable retention periods.

For Tutor mode, conversations are retained in the Moodle database and administrators can configure the period from 30 to 1,000 days, or disable automatic deletion. The cleanup task removes expired turns; a deletion request through Moodle privacy mechanisms remains the priority.

For Search, Creation and Administration modes, retention of indexes, drafts, histories and logs depends on enabled uses and service configuration.

Frequently asked questions

Answers to check with your IT department.

Where is Pimenko AI data hosted?

Service data is hosted in France on Scaleway infrastructure. The service data flows remain within the European Union, according to the contractual scope and deployed configuration.

Is the Pimenko AI plugin GDPR compliant?

Pimenko AI is designed with pseudonymisation, consent, configurable retention, and Moodle export and deletion mechanisms. Final compliance depends on the institution's configuration, contract and selected uses.

Is my data used to train AI models?

No. Within Pimenko AI's contractual scope, received data is used to process the request and operate the service, not to train a third-party model. This clause must be confirmed in the applicable terms or DPA.

Does the European AI Act apply to Pimenko AI?

The classification depends on the system's purpose and configuration. Pimenko AI is designed as a non-decision-making assistance tool: it does not determine access to training, grades, or learner rights. Legal validation remains recommended.

Which AI model does Pimenko AI use?

The service uses Mistral Open Weight models and a service layer developed by Pimenko.

How can I disable the plugin for particular courses or users?

The administrator controls mode activation and Moodle permissions. Course-level settings also apply the institution's pedagogical choices.

Who is responsible for data processing?

The institution generally defines the purposes and means of processing and acts as the data controller. Pimenko acts as a processor within the applicable contractual framework.

Can I use Pimenko AI with an on-premises hosted model?

This depends on the selected model, provider and architecture. It can be assessed with Pimenko where internal hosting or stronger control is required.

Get started with confidence

To return to the Moodle AI plugin, see the general Pimenko AI overview.

Have a question? Our team responds within 24 hours.