Mareante
Sign in
Mareante Compliance & Legal Architecture
Document URL: https://mareante.nuuno.art/legal/ai-transparency

Artificial Intelligence Transparency Notice

Transparency disclosure pursuant to Article 50 of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689).

Version 0.2.0-reviewLast updated: 2026-11-10

Table of Contents

Notice: Engineering compliance draft for Mareante (not yet reviewed by an independent attorney).

1. Statutory Context: EU Artificial Intelligence Act (Article 50)

This Transparency Notice is provided by Nuno Ribeiro, operating as a Swiss sole proprietorship (Einzelunternehmen) trading as "Mareante" ("Mareante"), in compliance with Article 50 of Regulation (EU) 2024/1689 (EU AI Act).

Mareante operates an autonomous software engineering and delivery platform. We are committed to transparency regarding the role, capabilities, and operational boundaries of artificial intelligence systems deployed across our services.

2. Disclosure of AI Interaction (Article 50(1) EU AI Act)

When you use the Mareante platform at mareante.nuuno.art:

  1. Direct AI Interaction: You interact directly with an artificial intelligence system. Both our text-based chat interface and our live interactive voice speech interface are driven by automated generative AI agents.
  2. Synthetic Conversational Speech: In voice sessions (FR-VC-01), the spoken audio you hear is real-time synthetic speech generated by an artificial intelligence model, not a human operator.
  3. No Impersonation: Mareante agents do not impersonate human engineers, support personnel, or legal advisors. At all times, the system identifies itself as an automated AI system.

Mareante integrates foundation models to understand technical requirements and generate software engineering artifacts:

  • AI Foundation Model Provider: Google Cloud / Google Ireland Limited (Google Gemini API).
  • Text & Code Generation: Google Gemini Flash and specialized reasoning models for technical requirement decomposition, task graph generation, Rhai tool extension synthesis, and test-driven code implementation.
  • Voice Synthesis & Processing: Google Gemini Live API utilizing bi-directional WebSockets for real-time natural language speech interpretation and low-latency audio synthesis.
  • Enterprise Data Protection: Our integration operates under commercial enterprise API terms ensuring that customer prompts, project source code, and live voice audio streams are never used to train or fine-tune Google's public foundation models.

4. Labelling of AI-Generated Content (Article 50(2) EU AI Act)

Mareante ensures that software artifacts, documentation, and commits generated by the platform are detectably identified:

  1. Commit Attribution: Source code commits created by platform agents are attributed with a standardized AI author signature (e.g., Co-authored-by: Mareante Agent <agent@mareante.nuuno.art>) in the git commit metadata.
  2. Pull Request & File Headers: Generated pull requests and newly scaffolded files embed standardized metadata stating that content was synthesized by Mareante AI agents.
  3. Machine-Readable Provenance: Deployed software repositories carry build manifests and hash-chained audit references (FR-AU-01) detailing the specific prompt versions, specification schemas, and model reviews that produced the release.

Autonomous agents in Mareante do not possess unrestricted authority. The platform architecture enforces strict human-in-the-loop governance:

[User Natural Language Prompt]
         │
         ▼
[AI Drafts Spec & Task Plan]
         │
         ▼
[Mandatory Human Approval Gate] ──(Approval Refused)──► [Plan Terminated]
         │ (Approved)
         ▼
[Sandboxed Docker Execution]
  - No host mounts
  - Resource limits (CPU/RAM)
  - Strict egress allow-list
  - Pre-allocated Budget Reservation
         │
         ▼
[Dual-Model Verification Review]
         │
         ▼
[CRITICAL GATE: Explicit Human UI Approval Required for Deploy / Secret / Budget]
         │
         ▼
[Production Deployment to Target]
  1. Sandboxed Execution (FR-EX-01): Agent code generation and test execution take place inside ephemeral, isolated Docker containers with zero host filesystem mounts, strict memory/CPU quotas, and capability-locked egress proxies.
  2. Firm Budget Caps (FR-BG-01–FR-BG-04): Every model interaction and compute task requires an advance budget reservation from the tenant's ledger. When a cap is reached, execution transitions to stopped_budget and halts.
  3. High-Risk Actions Exclusively Gated by Human UI Approval (FR-PL-04):
  4. Autonomous agents are technically barred from deploying code to production URLs or external servers without explicit, affirmative approval clicked in the Mareante web UI by an authorized user.
  5. Spoken approvals via voice are legally and technically insufficient for deployments, budget increases above project caps, domain purchases, or external credential integrations (FR-VC-05).
  6. Dual-Model Verification (FR-VR-02): Before presenting code for deploy approval, a second, independent model instance that did not generate the code reviews the diff against confirmed specification criteria.

Users and developers must take into account the following inherent boundaries:

  1. Probabilistic Reasoning: Generative language models predict tokens based on statistical patterns. They do not possess consciousness or common-sense human judgment.
  2. Potential for Hallucination: While dual-model verification (FR-VR-02) and test-driven execution (FR-EX-02) drastically reduce false outputs, language models can occasionally suggest obsolete libraries or suboptimal architectures.
  3. Security and License Verification: The platform scans dependencies and runs automated test suites (FR-VR-01), but the customer remains responsible for conducting human code audits and ensuring third-party open-source license compliance before deploying applications commercially.
  4. Non-Legal & Non-Financial Advice: The platform’s agents do not provide legal, financial, or certified engineering safety advice.

For questions regarding this AI Transparency Notice or our algorithmic governance practices:

  • Operator: Nuno Ribeiro (Mareante)
  • Email: [To be provided by operator: {{LEGAL_EMAIL}}]
  • DSA & AI Governance Contact: [To be provided by operator: {{DSA_CONTACT_EMAIL}}]