Regolo.ai inference API (EU-hosted open-weight models)

par Regolo.ai

AI Service

Faible confiance

Regolo.ai is a managed LLM inference API: customers call an OpenAI-compatible endpoint against third-party open-weight models (the site's top models are GLM 5.2, Qwen 3.5 122b, Faster Whisper Large v3, Gemma 4 31b and GPT OSS 120b) hosted on Seeweb's GPU cloud in Lazio and Lombardy, Italy, plus unspecified 'additional regions' from parent group DHH S.p.A. Customers can also deploy any vLLM-compatible Hugging Face model, including private ones, on dedicated GPUs. It does not itself make decisions about people; it processes whatever the customer sends, which may include personal data, documents, audio and images. Rated medium. In its favor: EU hosting, a GDPR Art. 28 DPA, stated content-level zero data retention (prompts and outputs never stored or reused for training), and infrastructure certifications. Cautions: the independent Zero Data Retention certificate (24 July 2026) is image-only, so its issuer and scope could not be read; the ISO and other badges are labeled datacenter certifications and may not cover the inference service; connection logs and usage metadata are kept until account deletion despite 'no records of any kind' wording; the sub-processor list is not published and the extra DHH regions and VAST Data's role are not specified; Seeweb disclaims being an AI Act provider and any responsibility for model content or security, leaving those with unnamed third-party model developers and the customer; and the Terms contradict themselves on Annex III high-risk uses (section 5.1 permits with care, 5.2 treats as a material breach) and bar integrating Regolo into products resold to third parties without a separate agreement. The rating would rise for high-risk or confidential workloads until the ZDR certificate, sub-processors and high-risk-use position are confirmed.