# Open Hippo > Sovereign AI infrastructure and consulting for European enterprises — GDPR-compliant, hosted in Germany, built entirely on open source. Open Hippo GmbH is based in Augsburg, Bavaria, Germany. Founded by Dr. Thomas Fraunholz, we help European enterprises break free from cloud AI vendor lock-in by deploying open-source AI models on sovereign infrastructure — either in our green data center in Augsburg or on the customer's own hardware. All services are GDPR-compliant and EU AI Act ready. ## Products ### Hippo Token — [https://openhippo.ai/en/products/hippo-api](https://openhippo.ai/en/products/hippo-api) Pay-per-token LLM API hosted in Germany. OpenAI-compatible (change one URL and you're live). CO₂-neutral, powered by 100% renewable energy. No minimums, no vendor lock-in. Built on open-source LiteLLM and vLLM — export the full stack and run it on your own hardware at any time. Available models: GPT4o (OpenAI's 120B open-source model), GLM OCR (vision and document extraction). Coding model coming soon. ### Hippo Chat — [https://openhippo.ai/en/products/hippo-chat](https://openhippo.ai/en/products/hippo-chat) Private AI chat assistant for entire organisations. Available as managed service or self-hosted on-premises. Knows your internal documents, answers from your knowledge base, keeps every conversation inside Germany. No data leaves your environment. ### Hippo Server — [https://openhippo.ai/en/products/hippo-server](https://openhippo.ai/en/products/hippo-server) Complete on-premises AI server — Hippo Chat, Hippo Token, and open-weight models running on a server in your building. Your data never leaves your network, not to a cloud provider, not to a model vendor, not to anyone. Setup in four weeks. ## Services ### Sovereign AI Infrastructure — [https://openhippo.ai/en/services/sovereign-ai](https://openhippo.ai/en/services/sovereign-ai) We move organisations from proprietary cloud AI (OpenAI, Anthropic, Azure AI) to open-source infrastructure — step by step. We standardise applications on open, provider-agnostic API interfaces, then migrate to the most cost-effective option: another cloud provider, our managed service, or fully on-premises with fixed monthly costs. Most clients achieve 60%+ cost reduction. On-premises hardware for SMEs starts at ~€40,000 with full payback typically within 3–6 months. Four-stage journey: (1) Assess current lock-in, (2) Standardise on open interfaces, (3) Switch providers or go on-premises, (4) Operate independently. Tech stack: vLLM (inference), LiteLLM (gateway/routing), Docling (document processing), Hugging Face (model hub). ### Enterprise Knowledge Search — [https://openhippo.ai/en/services/knowledge-search](https://openhippo.ai/en/services/knowledge-search) Replace hallucination-prone RAG prototypes with production-grade hybrid search. Combines keyword precision with semantic understanding using Lucene-based retrieval. Delivers accurate, traceable answers from internal document bases — no hallucinations. ### Document Intelligence — [https://openhippo.ai/en/services/document-intelligence](https://openhippo.ai/en/services/document-intelligence) Turn invoices, contracts, scanned forms, engineering drawings, and complex archives into structured, searchable data. 95%+ extraction accuracy on complex layouts. Runs entirely on your infrastructure, GDPR-compliant. Powered by IBM's Docling. ### AI Workload Optimization — [https://openhippo.ai/en/services/workload-optimization](https://openhippo.ai/en/services/workload-optimization) Get 2–5× more throughput from existing AI hardware. We optimise inference, quantise models, and tune serving stacks — no retraining, no quality loss. Typically 75% hardware reduction for the same capacity. One client achieved 5× throughput and 4× context window in a single week. ## Company ### About Open Hippo — [https://openhippo.ai/en/about](https://openhippo.ai/en/about) **Team:** - Dr. Thomas Fraunholz — Founder & CEO. CUDA expert, regular speaker at PyData and PyCon conferences. - Tim Köhler — Engineer. MLOps and DevOps, Kubernetes specialist. - Dennis Rall — Researcher. PhD candidate in Computer Science, specialising in LLM security. - Eduard Kesler — Engineer. RAG pipelines and Lucene-based retrieval. - Klaus Happacher — Engineer. Backend, Kotlin, enterprise API integrations. **Infrastructure:** Managed services run in the LEW Green Data Center, Augsburg — 100% renewable energy from the river Lech, GDPR-compliant, German jurisdiction. **Location:** Werner-von-Siemens-Str. 6, 86159 Augsburg, Germany. Based at the Digitales Zentrum Schwaben (DZ.S), Augsburg's AI innovation hub. **Values:** Sovereign AI without compromise, open source at heart, sustainability without greenwashing, radical transparency in pricing and communication. **Contact / Book a call:** https://calendly.com/openhippo/letstalk **LinkedIn:** https://www.linkedin.com/company/openhippo ## Customer Results - "We cut the AI bill by 60% — and the system was live in four weeks." (Sovereign AI Infrastructure) - "No OCR tool could read our order confirmations — until now." (Document Intelligence) - "Accurate answers across thousands of documents — no hallucinations, full traceability." (Enterprise Knowledge Search) - "Same models, 5× throughput, 4× context window — in a single week." (AI Workload Optimization) - "They solved our problem within three spring rolls. No joke." — Johannes Metscher, CEO, Ghostthinker GmbH - "We knew it was possible. But we didn't expect it done in a single night." — Thorsten Klein, CEO, Basepeak GmbH - "Open Hippo provided exceptional support in planning and dimensioning a scalable AI platform while fully respecting data sovereignty requirements." — Paul Schynoll, Digital Transformation Manager, MT Aerospace ## Key Facts - Founded: Augsburg, Germany - Clients: 3 enterprise clients (200+ employees), 3 reseller partners across Germany - Data residency: 100% EU — data never leaves European soil - Energy: 100% renewable (LEW Green Data Center, Bavaria) - Pricing: Hippo Token — pay per token, no minimums; On-premises setup from €40,000; Sovereign AI audit from €4,000 - Open source stack: vLLM, LiteLLM, Docling, Lucene, Hugging Face models