// FAQ

Frequently Asked Questions

Everything you need to know to start working with us.

01. How long does a project take?

It depends on the complexity. A website takes 4-8 weeks, an e-commerce 8-12 weeks, a web app 12-20 weeks.

02. What is the minimum budget?

We always assess every project carefully. If the budget isn't sufficient, thanks to our partner network we can help you find the missing funds through grants, financing or alternative solutions.

03. Do you offer post-launch support?

Yes, all projects include 30 days of free support. We also offer ongoing maintenance plans.

04. Do you work with international clients?

Absolutely. We work remotely with clients across Europe and beyond.

05. What is local AI and why should I choose it?

Local AI (or local-first AI) is artificial intelligence that runs entirely on the client's infrastructure — on-premise servers, edge boxes, dedicated hardware — without queries or data leaving the company perimeter. It's the right choice when you need to comply with GDPR on sensitive data, NDAs with clients, Public Administration sovereignty constraints, or simply when you don't want to depend on foreign cloud providers.

06. Is local AI really compliant with GDPR and the AI Act?

Yes, and with less complexity than cloud AI: the fact that no data leaves the perimeter simplifies the Records of Processing Activities, reduces the risk of extra-EU transfers and facilitates the DPIA. For 'high-risk' systems under the AI Act (EU 2024/1689), additional conformity assessments still apply, which we handle during the scoping phase.

07. Where is Omniproject based?

Omniproject has its operational headquarters in Rome, at Via Arenula 21. We operate throughout Italy with on-site projects (SMEs, PA) and remotely. For the Public Administration we are MePA-ready and accredited for AgID/CAD audits.

08. What is on-premise AI and why choose it in Italy?

On-premise AI is artificial intelligence that runs entirely on servers or edge hardware inside the organization's perimeter, without sending data to foreign cloud providers. In Italy it's the right choice when you process personal data under GDPR, information subject to professional secrecy or Public Administration data bound to national residency: no data crosses the border, simplifying the Records of Processing Activities and avoiding exposure to the US CLOUD Act. Omniproject designs and installs these systems in Rome and throughout Italy, on-site or remotely.

09. How much does a local RAG system cost?

The cost depends on four factors: volume and heterogeneity of the document sources, number of concurrent users, inference hardware (from an edge mini-PC to a GPU server) and integrations with existing systems. The logic is the opposite of the cloud: higher upfront investment but zero marginal cost per query — every subsequent query is free. This makes local RAG cost-effective above a certain volume threshold and for those who can't afford to let data leave the perimeter. We outline a concrete range in a free 30-minute discovery call.

10. Local AI or ChatGPT: which is better for the Public Administration?

For the Public Administration, local AI is almost always the mandatory choice. Cloud services like ChatGPT send data to extra-EU servers subject to the CLOUD Act, incompatible with data sovereignty constraints, AgID guidelines and the national residency required for many public processing activities. A local AI delivers the same conversational assistant (Q&A on documents, front-desk support, document search) while keeping data and inference inside the entity's perimeter. ChatGPT remains useful for public and non-sensitive activities; for citizens' data, local is needed.

11. Is local AI less powerful than cloud AI?

No, not for enterprise and Public Administration use cases. Open-weight models such as Qwen, Llama and Mistral, run with modern engines (vLLM, Ollama), reach quality suitable for document Q&A, automation and assistance, even running on accessible hardware thanks to 4-bit quantization. For specific tasks on proprietary data, a local model well integrated with RAG often outperforms a larger generic model that lacks access to your documents.

12. How long does it take to install a local AI?

A proof-of-concept on a circumscribed knowledge base typically takes 3-6 weeks: scoping, RAG pipeline, integration and testing. A production system — more sources, access controls, staff training — takes longer depending on the complexity. We always start from a narrow, measurable perimeter, so you see the value before scaling.

Didn't find the answer you were looking for?

$ Contact us →