https://qdna.fr/en/blog/ Sovereign AI platforms, local LLMs, GPU sizing, RAG and agents for enterprises. en Mon, 07 Sep 2026 08:00:00 +0000 qdna-site scripts/gen_feed.py https://qdna.fr/en/blog/ia-conforme-rgpd-liste-de-controle https://qdna.fr/en/blog/ia-conforme-rgpd-liste-de-controle Sun, 06 Sep 2026 08:00:00 +0000 GDPR-compliant AI: what the text requires, what the CNIL and the EDPB recommend, what the EU AI Act adds and when. Sourced checklist, chatbots, RAG and agents included. https://qdna.fr/en/blog/plateforme-ia-locale-llm-open-source-analyse https://qdna.fr/en/blog/plateforme-ia-locale-llm-open-source-analyse Sat, 05 Sep 2026 08:00:00 +0000 A local AI platform on an open-source LLM: confidentiality, permanence and control of the model secured, licence and threshold manageable. Benchmarks, licences and costs verified. https://qdna.fr/en/blog/deepseek-v4-flash-vision-exp https://qdna.fr/en/blog/deepseek-v4-flash-vision-exp Mon, 31 Aug 2026 08:00:00 +0000 DeepSeek released its first open-weight multimodal model on 31 August 2026: what the vision tower adds, and what the experimental label commits you to. https://qdna.fr/en/blog/qwen3-8-flash-next-mlx-mac-studio-m5-ultra https://qdna.fr/en/blog/qwen3-8-flash-next-mlx-mac-studio-m5-ultra Sun, 30 Aug 2026 08:00:00 +0000 180 billion parameters of which 6 active, a measured 111.6 GB in 4-bit MLX, 24 KiB of cache per token: the real sizing on a 512 GB M5 Ultra. https://qdna.fr/en/blog/glm-5-3-flash-mlx-mac-studio-m5-ultra https://qdna.fr/en/blog/glm-5-3-flash-mlx-mac-studio-m5-ultra Sun, 30 Aug 2026 08:00:00 +0000 A measured 334.1 GB in 8-bit MLX, 11 KiB of cache per token, a declared one-million-token window: the sizing of GLM-5.3-Flash on an M5 Ultra. https://qdna.fr/en/blog/debit-llm-conditions-de-mesure https://qdna.fr/en/blog/debit-llm-conditions-de-mesure Sun, 30 Aug 2026 08:00:00 +0000 The same model goes from 14 to 290 tokens per second depending on the machine, and software, compression and user count matter as much. Eight sourced tables, explained in plain language. https://qdna.fr/en/blog/traduction-juridique-ia-rgpd https://qdna.fr/en/blog/traduction-juridique-ia-rgpd Fri, 28 Aug 2026 08:00:00 +0000 Translating a contract online communicates it whole to a third party. Why secrecy is not cured by contract, and how to translate on-premises. https://qdna.fr/en/blog/patterns-rag-avances-entreprise https://qdna.fr/en/blog/patterns-rag-avances-entreprise Fri, 28 Aug 2026 08:00:00 +0000 Chunk with context, filter by clearance before scoring, decompose, return wide, measure recall: the five patterns that make a RAG system hold in production. https://qdna.fr/en/blog/moteur-orchestration-ia-routage-llm https://qdna.fr/en/blog/moteur-orchestration-ia-routage-llm Fri, 28 Aug 2026 08:00:00 +0000 Classify the request, pick the model, pass through a gateway with quotas: how an orchestration engine decides what stays on-premises and what leaves for an API. https://qdna.fr/en/blog/mac-studio-m5-ultra-llm-local https://qdna.fr/en/blog/mac-studio-m5-ultra-llm-local Fri, 28 Aug 2026 08:00:00 +0000 512GB at 1.2TB/s: what the M5 Ultra changes for a local LLM, which Flash models fit, and what MLX quantisation costs in quality, with measured figures. https://qdna.fr/en/blog/ia-donnees-sensibles-zero-retention https://qdna.fr/en/blog/ia-donnees-sensibles-zero-retention Fri, 28 Aug 2026 08:00:00 +0000 Zero retention is a contractual commitment, not an observable property. The four places data settles anyway, and what running on-premises changes. https://qdna.fr/en/blog/chatbot-ia-rgpd-transparence https://qdna.fr/en/blog/chatbot-ia-rgpd-transparence Fri, 28 Aug 2026 08:00:00 +0000 Inform, ground, limit, prove: the four obligations of an enterprise chatbot, and why three are decided in the architecture, not in the terms of service. https://qdna.fr/en/blog/asrock-ai-box-a395-qwen38-27b-strix-halo https://qdna.fr/en/blog/asrock-ai-box-a395-qwen38-27b-strix-halo Thu, 20 Aug 2026 08:00:00 +0000 The ASRock AI BOX-A395 (Ryzen AI Max+ 395, 128 GB shared LPDDR5X) runs multimodal Qwen3.8-27B at its native 262,144-token context, using 30% of memory. https://qdna.fr/en/blog/qwen3-8-27b-long-context-on-premises https://qdna.fr/en/blog/qwen3-8-27b-long-context-on-premises Tue, 18 Aug 2026 08:00:00 +0000 Gated DeltaNet hybrid architecture, a 64 KiB-per-token KV cache, native 262,144-token context: the real hardware sizing to serve Qwen3.8-27B on site. https://qdna.fr/en/blog/glm-5-3-deep-on-premise https://qdna.fr/en/blog/glm-5-3-deep-on-premise Tue, 18 Aug 2026 08:00:00 +0000 A full study of GLM 5.3: 744 billion parameters, 40 active, a one-million-token context, and measured throughput from DGX Spark to the GB300 NVL72 rack. https://qdna.fr/en/blog/modeles-open-weight-local-2026 https://qdna.fr/en/blog/modeles-open-weight-local-2026 Sun, 02 Aug 2026 08:00:00 +0000 The open-weight models that run on-premises in 2026: real size, active parameters, context window and the hardware each one needs, tier by tier. https://qdna.fr/en/blog/etude-de-cas-deepseek-dgx-station-pme https://qdna.fr/en/blog/etude-de-cas-deepseek-dgx-station-pme Sun, 02 Aug 2026 08:00:00 +0000 A sizing scenario, not a customer story: DeepSeek V4 Flash 0731 on a DGX Station GB300 for an 80-person SME. Estimated throughput, 3-year TCO, break-even against APIs. https://qdna.fr/en/blog/deepseek-local-dgx-station-spark-vllm https://qdna.fr/en/blog/deepseek-local-dgx-station-spark-vllm Sun, 02 Aug 2026 08:00:00 +0000 DeepSeek V4 Flash 0731 (284B, 13B active, 1M context) on DGX Spark, a 2× Spark cluster and DGX Station GB300: vLLM 0.25+ recipes, benchmarks and real cost. https://qdna.fr/en/blog/benchmark-independant-modeles-open-weight-2026 https://qdna.fr/en/blog/benchmark-independant-modeles-open-weight-2026 Sun, 02 Aug 2026 08:00:00 +0000 Vendor-published benchmark results for DeepSeek V4 Flash 0731 and GLM 5.2 (open weight, 2026): Terminal Bench, NL2Repo, GPQA Diamond, HLE. No estimates. https://qdna.fr/en/blog/benchmark-deepseek-v4-flash-0731-vs-opus5-vs-gpt55 https://qdna.fr/en/blog/benchmark-deepseek-v4-flash-0731-vs-opus5-vs-gpt55 Sun, 02 Aug 2026 08:00:00 +0000 DeepSeek V4 Flash 0731 (284B, 13B active, 1M context): vendor-published scores, real cost per token against the APIs, and what running on-premises changes. https://qdna.fr/en/blog/couche-semantique-ontologie-agents-ia https://qdna.fr/en/blog/couche-semantique-ontologie-agents-ia Thu, 23 Jul 2026 08:00:00 +0000 An ontology-based semantic layer gives AI agents governed context. Architecture, knowledge graph, eBPF enforcement and sovereign serverless deployment. https://qdna.fr/en/blog/qwen3-8-max-preview https://qdna.fr/en/blog/qwen3-8-max-preview Tue, 21 Jul 2026 08:00:00 +0000 Qwen3.8-Max: 2.4 trillion parameters, 95 billion active, and since August 2026 the base model weights published on Hugging Face. What it changes for local LLMs. https://qdna.fr/en/blog/kill-switch-llm-local https://qdna.fr/en/blog/kill-switch-llm-local Sun, 19 Jul 2026 08:00:00 +0000 June 2026: foreign access to two major AI models suspended by US order. Report No. 3054 proposes an open-source tax credit for SMEs. https://qdna.fr/en/blog/rag-entreprise-guide https://qdna.fr/en/blog/rag-entreprise-guide Sat, 18 Jul 2026 08:00:00 +0000 Enterprise RAG in 2026: chunking, hybrid search, reranking and access control. The architecture for connecting an LLM to your documents, on-premises. https://qdna.fr/en/blog/quel-llm-open-source-local-choisir https://qdna.fr/en/blog/quel-llm-open-source-local-choisir Sat, 18 Jul 2026 08:00:00 +0000 Comparison of the best open-source LLMs to run locally in 2026: GLM, DeepSeek, Kimi, Mistral, Qwen. Size, strengths, licence and NVIDIA, HPE, Dell hardware. https://qdna.fr/en/blog/prix-serveur-ia-guide https://qdna.fr/en/blog/prix-serveur-ia-guide Sat, 18 Jul 2026 08:00:00 +0000 How much does an AI server cost in 2026? Ranges by tier in euros and dollars, from €5,850 ($6,780) for a workstation to about €453,000 ($525,000) excl. VAT for an 8-GPU H200 server. https://qdna.fr/en/blog/orchestrer-plusieurs-agents-de-code https://qdna.fr/en/blog/orchestrer-plusieurs-agents-de-code Sat, 18 Jul 2026 08:00:00 +0000 Development is moving from the single agent to the fleet: a supervisor delegates to sub-agents, keeps context, and review remains the only real bottleneck. https://qdna.fr/en/blog/materiel-llm-prefill-decode-kv-cache https://qdna.fr/en/blog/materiel-llm-prefill-decode-kv-cache Sat, 18 Jul 2026 08:00:00 +0000 Prefill is compute-bound, decode is bandwidth-bound, the KV cache grows with every token. Which of the three decides your hardware, from PCIe card to cluster. https://qdna.fr/en/blog/installer-llm-local-entreprise https://qdna.fr/en/blog/installer-llm-local-entreprise Sat, 18 Jul 2026 08:00:00 +0000 Deploying a local LLM in the enterprise: model, NVIDIA hardware certified by HPE, Dell or Supermicro, vLLM or llama.cpp runtime, gateway and security. https://qdna.fr/en/blog/ia-generative-rgpd-souverainete https://qdna.fr/en/blog/ia-generative-rgpd-souverainete Sat, 18 Jul 2026 08:00:00 +0000 Generative AI and GDPR in 2026: what the CNIL recommends, the extraterritorial risk, and how local AI puts data control back in the hands of the enterprise. https://qdna.fr/en/blog/ia-en-local-sur-pc https://qdna.fr/en/blog/ia-en-local-sur-pc Sat, 18 Jul 2026 08:00:00 +0000 Two ways to run AI on your own machine in 2026: free with open-weight models, or a DGX Spark mini-supercomputer (ASUS, Dell, Lenovo, MSI). https://qdna.fr/en/blog/open-knowledge-format-memoire-agents https://qdna.fr/en/blog/open-knowledge-format-memoire-agents Fri, 17 Jul 2026 08:00:00 +0000 OKF represents in Markdown the knowledge AI agents consume. Portable, versionable and free of lock-in, it serves as the memory layer of sovereign platforms. https://qdna.fr/en/blog/ia-applications-metier https://qdna.fr/en/blog/ia-applications-metier Thu, 16 Jul 2026 08:00:00 +0000 A single internal AI API serves CRM, support, ERP and document management, with virtual keys, budgets and telemetry per application. https://qdna.fr/en/blog/coder-avec-ia-modele-local https://qdna.fr/en/blog/coder-avec-ia-modele-local Thu, 16 Jul 2026 08:00:00 +0000 An OpenCode harness, open-weight models served by vLLM, a LiteLLM gateway with quotas: AI coding help without a line of source leaving your infrastructure. https://qdna.fr/en/blog/chatbot-ia-souverain-entreprise https://qdna.fr/en/blog/chatbot-ia-souverain-entreprise Thu, 16 Jul 2026 08:00:00 +0000 The enterprise chatbot use case: an internal ChatGPT connected to your documents through RAG, GDPR-compliant and hosted on-premises. Hardware by segment. https://qdna.fr/en/blog/qu-est-ce-qu-une-plateforme-ia-souveraine https://qdna.fr/en/blog/qu-est-ce-qu-une-plateforme-ia-souveraine Sun, 05 Jul 2026 08:00:00 +0000 A sovereign AI platform, also called on-premise AI, hosts models, data, and logs on your own hardware, on-premises or hybrid, under GDPR and HDS compliance. https://qdna.fr/en/blog/pieges-mise-en-production-llm-sur-site https://qdna.fr/en/blog/pieges-mise-en-production-llm-sur-site Sun, 05 Jul 2026 08:00:00 +0000 Eight concrete pitfalls of deploying an on-premises LLM inference service into production, and the measures that prevent each one. https://qdna.fr/en/blog/on-premise-ou-api-cout-llm https://qdna.fr/en/blog/on-premise-ou-api-cout-llm Sun, 05 Jul 2026 08:00:00 +0000 We identify the LLM cost break-even point between on-premises and pay-per-token, with the role of the KV cache and sovereign hosting. https://qdna.fr/en/blog/de-serveur-gpu-a-service-llm-production https://qdna.fr/en/blog/de-serveur-gpu-a-service-llm-production Sun, 05 Jul 2026 08:00:00 +0000 A seven-phase method for moving from a GPU server to an LLM service in production, with a deliverable and an exit criterion for each phase. https://qdna.fr/en/blog/chatbot-vers-plateforme-agentique https://qdna.fr/en/blog/chatbot-vers-plateforme-agentique Sun, 05 Jul 2026 08:00:00 +0000 The harness, three-layer memory and skills turn a chatbot into an agentic platform that executes tasks and capitalises on procedures.