Reconectando con el servidor

Mantén esta pestaña abierta: tu trabajo no se ha perdido.

No hemos podido reconectar

Recarga la página para continuar donde lo dejaste.

La sesión ha caducado

Vuelve a cargar la página para iniciar sesión de nuevo.

English
ES EspañolPT PortuguêsEN English
Free assessment· 1h
Spain's authority is now operating and tick-box channels are under review. Would yours hold up?Act now →
AI agents — DSA SmartAgent ›

DSA SmartAgent

AI agents running on the platform's own infrastructure

An AI assistant applied to compliance is only usable if it meets three conditions: the data never leaves the system, it cites the document it relies on, and the decision stays with an identifiable person. All three are legal requirements, not technical preferences.

7 screens of the programme
app.dsacompliance.net DSA SmartAgent
01A catalogue of agents by program
02The agent's identity and system prompt
03Model, parameters and local inference
04Versioned prompt templates
05A document base isolated per entity
06The agent open inside the file
07An analysis citing the rule, and an unsigned draft
DSA Compliance

Screens from DSA Compliance v6.2 in a demo environment. All data shown is fictitious.

The obligation, precisely

What the rule requires, and in which article

For information only, not legal advice. Always check against the consolidated text in force.

Art. 14 Regulation (EU) 2024/1689
Effective human oversight of high-risk systems, with the ability to interpret the output and to decide not to use it.
Art. 22 GDPR
The right not to be subject to decisions based solely on automated processing that produce legal effects or similarly significantly affect the person.
Art. 26 Regulation (EU) 2024/1689
The deployer's obligations: use in line with the instructions, oversight by competent people, and retention of logs.
Arts. 28 and 32 GDPR
If the model ran at a third party there would be a processor relationship and a transfer. Local inference avoids both.
Penalty regime

Improper use of AI is penalised under whichever rule governs the processing: up to €35,000,000 or 7% for prohibited practices under art. 5 of the AI Act, and up to 4% for GDPR infringements.

Rules it covers
EU AI ActUE 2024/1689ISO 42001AI managementRGPD art. 22Human decision

The badges identify the obligation the program digitises. They are not certifications or conformity accreditations.

What it leaves as evidence

What gets handed over when somebody asks

  • A log of every run with model, prompt version, date and user
  • The documents the agent consulted for each answer
  • A document base isolated per entity, with no mixing between tenants
  • Generated drafts flagged as unsigned until reviewed
  • A version history of the system prompts
Who it applies to
Compliance teams with high case volumes
Departments short on expert judgement and heavy on repetitive work
Organisations that cannot send data to external AI services
Companies already required to document art. 14 human oversight
Frequently asked questions

What we get asked about SmartAgent

Does data leave for an external service?

No. Inference runs on the platform's own infrastructure, with the document base isolated per entity.

Does the agent decide?

No. It proposes and cites the source. Classification, signature and closure stay with the responsible person, and that's what gets recorded.

What if the agent gets it wrong?

There's a record of the model, the prompt and the documents consulted, which lets you reconstruct the answer and fix either the document base or the instruction.

Vea SmartAgent running with their cases

A 30-minute guided demo on the real platform, or a one-hour assessment session with a consultant.

← The 22 programmes of the platform