# EZOS.Hosting

> Easy Open Source Hosting since 2002.

EZOS.Hosting provides Managed Local AI on the customer's own GPU with Ollama and Open WebUI, open-source hosting with CyberPanel, and domain services. The company has offered hosting and support since 2002. Client accounts, the knowledgebase, and ordering are available through the HostBill portal at support.ezoshosting.com.

## Main pages

- [EZOS.Hosting](https://www.ezoshosting.com/): Overview of Managed Local AI, open-source hosting, domains, and support.
- [Managed Local AI](https://www.ezoshosting.com/managed-local-ai/): Managed Ollama-based local AI on the customer's own GPU with Open WebUI as the default interface.
- [AI Apps](https://www.ezoshosting.com/ai-apps/): Private knowledge, team chat, workflow automation, and creative GPU app options.
- [GPU Infrastructure](https://www.ezoshosting.com/gpu-infrastructure/): GPU server planning for local inference, image workflows, and private automation stacks.
- [Open Source Hosting](https://www.ezoshosting.com/open-source-hosting/): Managed open-source hosting with CyberPanel, SSL, DNS, backups, and support.
- [Domains](https://www.ezoshosting.com/domains/): Domain registration, renewal, transfer guidance, and DNS support.
- [Support](https://www.ezoshosting.com/support/): Support routes for the knowledgebase, tickets, and managed setup requests.
- [Contact](https://www.ezoshosting.com/contact-us/): Contact and intake paths for local AI, hosting, domains, and migrations.

## HostBill portal

- [Client area](https://support.ezoshosting.com/clientarea/): Secure client account and login area.
- [Knowledgebase](https://support.ezoshosting.com/knowledgebase/): Public support articles and common answers.
- [Cart](https://support.ezoshosting.com/cart/): Secure ordering portal for available services.

## Current benchmark and support signals

- Local API bridge benchmark: https://www.ezoshosting.com/managed-local-ai/#local-ai-api-bridge - OpenAI-compatible local endpoint readiness is benchmarked before any customer workflow depends on it.
- Private local agent safety benchmark: https://www.ezoshosting.com/managed-local-ai/#local-agent-safety-benchmark - includes the gpt-oss-safeguard 20B policy benchmark, OpenAI safeguard announcement, and Hugging Face safeguard-20b source checks.
- Support and contact benchmark intake: https://www.ezoshosting.com/contact-us/#benchmark-intake - request scoped fit checks for Team RAG, Business Secure, BYO Server Management, and Managed Local AI.
- Legacy contact recovery: https://www.ezoshosting.com/contact-us/ - recovered legacy contact and sales paths now route buyers to a current intake page instead of 404 pages.
- Resource hub: https://www.ezoshosting.com/blog/
- Private RAG readiness checklist: https://www.ezoshosting.com/blog/private-rag-readiness/ - checks document shape, access boundaries, retrieval risk, and go-live scope.
- Private code assistant benchmark: https://www.ezoshosting.com/blog/private-code-assistant-benchmark/ - validates private coding assistant fit before promising live team use.
- Private document intake benchmark: https://www.ezoshosting.com/blog/private-document-intake-benchmark/ - tests OCR, parsing, metadata, and source handling before RAG rollout.
- Visual RAG benchmark: https://www.ezoshosting.com/managed-local-ai/#local-ai-offer-tracks - Qwen3-VL-Embedding-8B and Qwen3-VL-Reranker-8B are current visual retrieval benchmark candidates for documents, screenshots, and diagrams.
- RAG retrieval quality audit: https://www.ezoshosting.com/managed-local-ai/#rag-retrieval-quality-audit - Qwen3-Embedding 0.6B/4B and Qwen3-Reranker 0.6B/4B are evaluated for source-grounded retrieval quality before production use.
- One-time Team RAG Retrieval Quality Audit (€431 one-time): https://support.ezoshosting.com/?cmd=cart&action=add&id=926 - bounded corpus, fixed questions, citation review, and a go/no-go report before recurring Team RAG.
- Workflow RAG/agent benchmark evidence (2026-08-02): qwen3-embedding:0.6b plus gpt-oss:20b passed a fixed public EZOS Team RAG/agent benchmark with 0.8 top-1 retrieval accuracy and 3/3 grounded answer plus agent-boundary cases. This is public-corpus benchmark evidence only, not a customer-corpus production claim.
- Visual/document benchmark evidence (2026-08-02): qwen3-vl:8b passed a fixed public synthetic EZOS visual/document benchmark with 3/3 cases: invoice OCR, Team RAG screenshot interpretation, and local runtime-gate panel reading. This is synthetic-public-sample benchmark evidence only, not a customer-corpus or production performance claim.
- Source links for current benchmark candidates: the Managed Local AI page links to Qwen3 Embedding, Qwen3-VL-8B, Ollama Qwen3-Coder, Gemma 4, and gpt-oss-20b references used for offer framing.

## Model-fit notes for the RTX 4000 Ada server

- Runtime status verified 2026-08-02: the NVIDIA RTX 4000 SFF Ada (20 GB), Ollama 0.17.7, and a representative gpt-oss:20b API smoke passed without disrupting running services. Customer workflows remain benchmark-first until their own corpus and task tests pass.
- Qwen3-Reranker is tracked for retrieval-quality audits, especially paired with Qwen3-Embedding 0.6B/4B on private knowledge bases.
- Qwen3-VL 8B + Docling/OCR is tracked for visual and scanned-document RAG intake before customer rollout.
- Standard model-fit boundary: start with Qwen3 8B for latency/concurrency or quantized Qwen3 14B for richer private assistant and RAG work. Qwen3-Coder 30B and other 30B/32B paths remain experimental on 20 GB VRAM, not standard offer capacity.
- Gemma 4 market refresh: gemma4:26b and gemma4:31b remain benchmark-only candidates until local smoke tests prove practical throughput.