The AI assistant that keeps your data in Spain
Brings the strongest open-source models under one chat interface — Gemma, Qwen, Nemotron, GLM, and whatever lands next — running on dedicated NVIDIA HGX B200 infrastructure on Spanish soil. Native voice in Spanish, English and the Spanish co-official languages. AES-256 encryption, GDPR by design, no training on your conversations.
OdiModel · chat
Summarise the five questions the NIS2 regulator is putting on the table this quarter.
European regulators are looking at five very concrete fronts in the AI stacks deployed by an essential operator:
- · Data residency and traceability across the inference lifecycle.
- · Supply chain: who you subcontract the compute to and where it lives.
- · Incident management and notification windows under NIS2.
Type or dictate your question…
ENTER to sendOpen catalogue
The best open-source models, under one interface
Each model optimised for a specific job — general text, reasoning, code, vision, agents. If you need one that's not on the list, we deploy it in 24 hours.
Native voice, official languages
The only assistant with full coverage of Spain's languages
Conversation and dictation in Spanish, English, Catalan, Basque and Galician. Built for regional administrations, public services and multilingual media.
Sovereignty by default
Your data never leaves Spain
Dedicated HGX B200 in a Spanish datacenter, AES-256 end-to-end, GDPR native, ENS, ISO 27001. No international transfers, no use for training.
Catalogue · 2026
Open list · expanding
Gemma-4-E4B
4B
Reasoning, code and general text. Light default model for everyday tasks.
Qwen 3.6
27B
Recently added. Extended reasoning and multilingual conversation. Strong size-to-latency balance for daily use.
Qwen 3 / 3.5
14B / 9B
Qwen family for general conversation and multilingual analysis — lighter alternatives to the 27B.
Nemotron-VL
8B
Vision-language: OCR, image analysis, information extraction from scanned documents or screenshots.
Nemotron
30B
Complex reasoning, mathematics and agentic tasks with longer reasoning chains.
GLM-5.1-FP8
754B
Reasoning, code and large-scale agentic tasks. The frontier model in the catalogue.
Models on request. If you need one that isn't on the list, we deploy it within 24 hours provided it meets open-source conditions.
Built-in capabilities
One interface, everything a team needs in a single session
01 · Chat
Persistent history
Conversations grouped by day, archive, manual delete. Accessible only by the account holder.
02 · Voice
Conversation and dictation
Two native modes: talk like a phone call, or dictate and have answers read aloud. Co-official language support.
03 · Web search
Live information
Up-to-date queries with source citation inside the conversation. Toggle per message.
04 · Documents
Upload and analyse
Documents and images for vision-language models. Library with search and delete at any time.
05 · Usage panel
Tokens and consumption
Conversations, messages and tokens broken down by model, with filters by day, week and month.
The stack underneath
Every layer chosen so the data never crosses a border
OdiModel runs on the same sovereign stack as the rest of GPU Solutions' products. The difference with a commercial assistant isn't the chat experience — it's what happens the moment you press send.
Compute
NVIDIA DGX B200
Nodes of 8 × B200 on HGX boards, dedicated to the customer, no GPU sharing.
Encryption
AES-256 end-to-end
In transit and at rest. Keys managed inside the EU.
Residency
Datacenters in Spain
EU data residency. Low latency for Iberian and southern European users.
Compliance
GDPR · ENS · ISO 27001
A Spanish company under European law. Native compliance, not a premium add-on.
Isolation
Per account, per session
Quotas, storage and history isolated. What one customer writes never touches another's context.
Sign in today · free during the beta
Try it with your own questions
Open registration, no commitment. Custom enterprise plan tailored to each partner. Direct support from the technical team within 24 business hours.
Your AI deserves real infrastructure.
Come see it. We'll invite you to the datacenter in Madrid. No PowerPoints.