Main Page: Difference between revisions
(Veille tech 2026-09-08: 1 ajouts GitHub trending) |
|||
| Line 171: | Line 171: | ||
= Nouveautés = | = Nouveautés = | ||
== Nouveautés implémentés | == Nouveautés implémentés == | ||
* [https://github.com/ynotopec/tricoteuses-k8s '''tricoteuses-k8s'''] (2026-09-08) : Kubernetes deployment for tricoteuses-juridique | * [https://github.com/ynotopec/tricoteuses-k8s '''tricoteuses-k8s'''] (2026-09-08) : Kubernetes deployment for tricoteuses-juridique | ||
| Line 244: | Line 242: | ||
---- | ---- | ||
= Priorités = | = Priorités = | ||
Revision as of 21:35, 8 September 2026
infocepo.com – Cloud, AI & Labs
Bienvenue sur le portail infocepo.com.
Ce wiki documente l’écosystème Cloud, IA, automatisation et lab d’Infocepo. Il s’adresse aux :
- administrateurs systèmes,
- ingénieurs cloud,
- développeurs,
- étudiants,
- curieux qui veulent apprendre en pratiquant.
L’objectif est simple : transformer la théorie en scripts réutilisables, schémas, architectures, APIs et laboratoires concrets.
Accès rapide
Portail principal
Assistant IA
Liste des pages du wiki
Vue d’ensemble
Démarrer rapidement
Parcours recommandés
- 1. Construire un assistant IA privé
- Déployer une stack type Hermes WebUI + Ollama + GPU
- Ajouter un modèle de chat et un modèle de résumé
- Brancher des données internes via RAG + embeddings
- 2. Lancer un lab cloud
- Créer un petit cluster Kubernetes, OpenStack ou bare-metal
- Mettre en place un pipeline de déploiement (Helm, Ansible, Terraform…)
- Ajouter un service IA : transcription, résumé, chatbot, OCR…
- 3. Préparer un audit ou une migration
- Inventorier les serveurs avec ServerDiff.sh
- Concevoir l’architecture cible
- Automatiser la migration avec des scripts reproductibles
Vue d’ensemble du contenu
- Guides IA & outils : assistants, modèles, évaluation, GPU, RAG
- Cloud & infrastructure : Kubernetes, OpenStack, HA, HPC, DevSecOps
- Labs & scripts : audit, migration, automatisation
- Comparatifs : Kubernetes vs OpenStack vs AWS vs bare-metal, etc.
Vision
Le but à long terme est de construire un environnement où :
- les assistants IA privés accélèrent la production,
- les tâches répétitives sont automatisées,
- les déploiements sont industrialisés,
- l’infrastructure reste compréhensible, portable et réutilisable.
Chaque sortie IA est mesurée en temps réel, les modèles faibles sont remplacés automatiquement, et les tâches sans valeur sont coupées. L'autonomie vient de mécanismes techniques qui fonctionnent quand personne ne regarde — pas de règles écrites dans un doc.
Catalogue rapide des services
| Catégorie | Service | Rôle |
|---|---|---|
| API | LLM | Modèles de chat, code, RAG, OCR |
| API | STT | Transcription audio |
| API | TTS | Synthèse vocale |
| API | realtime-ai | Temps réel WebSocket / WebRTC |
| API | IMAGE2TXT | OCR / VLM via endpoint dédié |
| API | summary | Résumé de textes longs |
| API | EMBEDDINGS | Embeddings pour RAG |
| API | ChromaDB | Base vecteur |
| API | TXT2IMAGE | Génération d’images |
| API | diarization | Segmentation locuteurs |
| Observabilité | monitoring | Dashboards techniques |
| Observabilité | status | Disponibilité des services |
| Observabilité | web-stat | Statistiques web |
| Observabilité | LLM-stat | Vue API / usage |
| Outils | dataLab | Environnement de travail hors-production |
| Outils | realtime translation | Traduction |
| Outils | Demos | Démonstrateurs |
Sélection AI & architecture par couche
Le tableau de bord Sélection AI présente l'architecture par couche des modèles IA déployés :
| Couche | Modèle | Alias(s) | Taille |
|---|---|---|---|
| HYBRID-CLOUD | Qwen3.6-35B | flash, default, thinking, vision | Option 135B MoE / 3B actif |
| Qwen3.8-Flash-Next | flash, default, thinking, vision, max | Option 2125B MoE, 51B n-grames internes, 6B actif, 1M contexte | |
| MiniMax-H3 | video | 33B dense | |
| BGE-M3 | embedding | 568M | |
| BGE-reranker-v2-m3 | reranker | 568M | |
| FLUX.2-klein-4B | image | 4B | |
| GLiNER2.5-Multi-V1 | privacy | 287M, mDeBERTa-v3-base — privacy filter: api-llm-privacy-proxy-gliner2 | |
| Public-Cloud | GLM-5.3-Flash | max | 320B MoE / 18B actif — 1M contexte, multimodal |
| LOCAL | Qwen3.5-9B | flash, default, thinking, vision | Option 1 9B dense |
| Qwen3.8-27B | flash, default, thinking, vision, max | Option 2 27B dense |
Frameworks
| Outil | Usage |
|---|---|
| Hermes | Usage général, raisonnement |
| OpenCode | DevSecOps |
| Kilo Code | Développement pur |
Mémoire
Pour la couche d'embedding vectoriel (RAG, mémoire sémantique) :
- BGE-M3 — 568M paramètres, architecture XLM-RoBERTa. Multi-fonction (dense + sparse + multi-vector), 100+ langues, licence MIT. Idéal comme moteur de mémoire à échelle.
Nouveautés
Nouveautés implémentés
- tricoteuses-k8s (2026-09-08) : Kubernetes deployment for tricoteuses-juridique
- dashboard-superset-mcp (2026-09-07) : Automate Apache Superset dashboard creation via MCP server 6.1.0+
- api-txt2image (2026-09-02) : Text-to-image generation API for AI-powered image creation
- omnivoice-tts (2026-09-01) : omnivoice-tts
- superset-k8s (2026-09-01) : superset-k8s
- dind-build (2026-08-31) : Docker-in-Docker build factory on Kubernetes — build, push, pull via K8s DinD pod
- quality-gate (2026-08-28) : quality-gate
- agent-saas (2026-08-27) : SaaS AI agent platform — deploy autonomous AI agents as a service
- api-convert2md (2026-08-26) : Document-to-Markdown conversion API — clean markdown from various formats
- qwen36-sglang (2026-08-24) : Qwen 3.6 deployment with SGLang for high-performance inference
- trafilatura-local (2026-08-23) : trafilatura-local
- api-rag (2026-08-19) : RAG API — semantic search over knowledge bases with retrieval-augmented generation
- models-todo (2026-08-19) : models-todo
- minimax-music3 (2026-08-18) : minimax-music3
- hermes-img-gen-infocepo (2026-08-15) : hermes-img-gen-infocepo
- ask-multiple-pdfs (2026-08-14) : A Langchain app that allows you to chat with multiple PDFs
- openclaw-install (2026-08-14) : OpenClaw installer — setup script for OpenClaw AI assistant deployment
- api-llm-custom (2026-08-13) : Custom LLM proxy API — route requests to any language model backend
- infocepo-infra-mcp (2026-08-12) : infocepo-infra-mcp
- api-mcp-openai (2026-08-11) : AI MCP OpenAI Integration
- api-embedding (2026-08-07) : Text embedding API — convert text to vector embeddings for semantic search
- api-reranker (2026-08-06) : Cross-encoder reranking API for improving search retrieval quality
- nemotron-embed (2026-07-31) : nemotron-embed
- coder-brain (2026-07-28) : AI-powered code analysis and generation engine
- translate-rt (2026-07-28) : Real-time translation API — multilingual translation with low latency
- qwen3.5-fp8-vllm (2026-07-27) : Qwen 3.5 FP8 vLLM model deployment
- ai-assistant (2026-07-17) : General-purpose AI assistant with task execution capabilities
- ai-security (2026-07-17) : AI security tools — audit and harden LLM deployments against common attacks
- api-audio2txt (2026-07-17) : Fast audio-to-text transcription API with real-time Whisper model support
- api-audio2txt-gemma4 (2026-07-17) : Gemma 4-powered audio transcription API for speech-to-text conversion
- api-audio2txt-qwen3 (2026-07-17) : Qwen 3-based audio transcription API leveraging advanced language model capabilities
- api-diarization (2026-07-17) : Speaker diarization API — identifies and separates multiple speakers in audio streams
- api-image (2026-07-17) : Image processing API — resize, convert, analyze, and transform images
- api-lightrag (2026-07-17) : Lightweight RAG (Retrieval-Augmented Generation) API for document-based Q&A
- api-llm-privacy-proxy (2026-07-17) : LLM privacy proxy — filter secrets and sensitive data from API calls
- api-llm-privacy-proxy-gliner2 (2026-07-17) : GLiNER2-powered LLM privacy proxy for named entity detection and filtering
- api-llm-privacy-proxy-openmed (2026-07-17) : OpenMED-powered LLM privacy proxy for medical text privacy filtering
- api-ocr (2026-07-17) : Document OCR API — extract text, tables, and structure from scanned documents
- api-realtime-ai (2026-07-17) : Real-time AI inference API with low-latency streaming responses
- api-tts-omnivoice (2026-07-17) : Text-to-speech API powered by OmniVoice for natural voice synthesis
- api-txt2audio (2026-07-17) : Text-to-audio generation API for converting text content to audio output
- artificial-human (2026-07-17) : Artificial human simulation — explore the boundary between human and AI behavior
- audio2txt (2026-07-17) : audio2txt
- common-good-accelerator (2026-07-17) : Platform for accelerating common good and social impact initiatives
- dgx-memory-fix (2026-07-17) : NVIDIA DGX memory management — diagnose and fix memory leaks on GPU systems
- diff-multi (2026-07-17) : Multi-format diff tool — compare files, directories, and structured data
- docker2rootless (2026-07-17) : docker2rootless
- evaluate-my-code (2026-07-17) : Code evaluation and quality assessment tool — automated peer review
- github-k8s-example (2026-07-17) : Kubernetes deployment example for GitHub Actions CI/CD pipelines
- gpu-cluster (2026-07-17) : GPU cluster management — k8s-based AI compute cluster with NVIDIA GPUs
- image-editor (2026-07-17) : AI image editor — modify, enhance, and generate images with AI
- k8s-domain-migrate (2026-07-17) : Kubernetes domain migration tool — move services between clusters
- k8s-internet-infrastructure (2026-07-17) : Internet infrastructure on Kubernetes — DNS, networking, and routing
- langextract-demo (2026-07-17) : Demo application showcasing langextract for LLM output extraction
- langfuse-k8s (2026-07-17) : Langfuse observability deployment on Kubernetes for LLM tracing and monitoring
- legal-agent (2026-07-17) : Legal AI agent — assist with legal document analysis and research
- litellm-proxy (2026-07-17) : LiteLLM proxy — unified API interface for multiple LLM providers
- llmbench (2026-07-17) : LLM benchmarking suite — measure and compare language model performance
- mail-search (2026-07-17) : Email search tool — query and analyze inbox contents
- mediawiki-migrate (2026-07-17) : MediaWiki migration tool — migrate and sync wiki content between platforms
- opencode-demo (2026-07-17) : OpenCode CLI demo — open-source alternative for AI-assisted coding
- opencode-glm-example (2026-07-17) : OpenCode example with GLM model integration
- pageindex-mvp (2026-07-17) : PageIndex MVP — semantic document search and indexing platform
- python2helm (2026-07-17) : Convert Python projects to Helm charts for Kubernetes deployment
- realtime-transcription (2026-07-17) : Real-time audio transcription service — live speech-to-text
- sh (2026-07-17) : sh
- vpn-network (2026-07-17) : VPN network configuration — secure remote access and networking
Priorités
Top tasks
- magnitude (2026-09-08) : Open source inference server for local models — works with Pi, OpenCode, Hermes, Codex
- minimind (2026-09-08) : Train a 64M-parameter LLM from scratch in 2h — minimal LLM training from scratch
- timesfm (2026-09-08) : Time Series Foundation Model by Google Research — time-series forecasting
- OpenMAIC (2026-09-08) : Open Multi-Agent Interactive Classroom — immersive multi-agent learning experience
- VoiceStudio (2026-09-08) : Open-source fully-local ElevenLabs alternative — voice cloning, dubbing in 646 languages
- feynman (2026-09-08) : Open source AI research agent — automated research and analysis
- ECC (2026-09-08) : Agent harness performance optimization — skills, memory, security for Claude Code, Codex
- Ajouter Presidio : anonymisation / masquage PII, socle RGPD.
- Ajouter llm-d : blueprints + charts Kubernetes pour industrialiser les déploiements.
- Ajouter Dynamo : orchestration inférence multi-nœuds.
- Ajouter GuideLLM : capacity planning / benchmark réaliste.
- Ajouter NeMo Guardrails : garde-fous et politiques.
- Coût unitaire par tâche : chaque API doit exposer son prix/ml/token en temps réel — pas une facture mensuelle.
- Qualité auto : précision/hallucination rate branchés sur les sorties IA, avec rollback automatique si qualité < seuil.
Backlog / Veille Technologique
Agents IA & Orchestration
- meetily — Assistant IA de reunion local avec transcription Parakeet/Whisper en direct, diarisation et resume Ollama. 100% local, aucun cloud.
- langflow (2026-09-08) : Builder et déployer des agents IA et workflows. UI visuelle low-code (154,000⭐)
- DesktopCommanderMCP — Serveur MCP pour Claude : controle terminal, recherche fichiers et edition de diffs.
- codex-plugin-cc — Utiliser Codex depuis Claude Code pour reviewer du code ou deleguer des taches.
- CubeSandbox — Sandbox instantanee, concurrente, securisee et leger pour agents IA.
- herdr — Multiplexeur d'agents qui vit dans votre terminal.
- claude-video — Donner a Claude la capacite de regarder des vidéos : download, extraction frames, transcription.
- OfficeCLI — Suite Office construite pour agents IA : lire, modifier, automatiser Word, Excel, PowerPoint en binaire unique.
- archify — Skill agent : generer de beaux diagrammes d'architecture avec export PNG/JPEG/WebP/SVG.
- claude-skills — 345 skills et plugins pour agents de codage (30+ agents, 70+ commandes, 330+ skills) pour Claude Code, Codex, Gemini CLI, Cursor, et 8+.
- chrome-devtools-mcp — Chrome DevTools pour agents de codage via MCP.
- Paperclip — Orchestrateur open-source pour coordonner et superviser une équipe d'agents IA autonomes
- OpenClaw
- codegraph — Indexation locale des graphes de code pour agents (Hermes, Claude, Codex sync auto).
- Understand-Anything — Transforme tout code en graphe de connaissances interactif explorable et questionnable.
- OpenHands — Agent IA autonome pour le développement logiciel
- Dify — Plateforme de développement d'applications IA (LLM Ops)
- browser-use — Framework pour contrôler les navigateurs via des agents IA
- LangChain — Framework pour applications basées sur les LLM
- FlowiseAI — Build LLM apps visually
- Rasa — Framework open-source pour chatbots et assistants vocaux
- codebase-memory-mcp - Serveur MCP haute performance pour l'indexation de codebases en graphe de connaissances persistant.
- Agent-Reach - Donnez des yeux a vos agents IA pour naviguer le web entier (Twitter, Reddit, YouTube, GitHub) en CLI, sans frais API.
- SkillSpector - Scanner de securite pour les competences d'agents IA : detecte vulnerabilites, patterns malveillants et risques.
- flue - Framework sandbox pour agents IA -- isolation et execution securisee de plugins/agents.
- agent-skills - Skills d'ingenierie de qualite production pour les agents de codage IA (Claude, Cursor, etc.).
- deer-flow - SuperAgent open-source a horizon long : recherche, codage et creation avec sandboxes, memoires, outils, sous-agents et passerelle de messages.
- Orca - ADE pour travailler avec une flotte d'agents paralleles. Lancez n'importe quel agent de codage avec votre propre abonnement, disponible sur desktop et mobile.
- agent-toolkit-for-aws - Serveurs MCP, competences et plugins officiels supports par AWS pour aider les agents IA a construire sur AWS.
- cognee - Plateforme de memoire IA open-source pour agents. Offrez a vos agents une memoire a long terme persistante via un moteur de graphe de connaissances auto-hebergt.
- page-agent - Agent GUI en-page JavaScript. Controlez les interfaces web en langage naturel.
- agent-native - Framework pour construire des applications natives pour agents IA.
- hallmark - Competence design anti-AI-slop pour Claude Code, Cursor et Codex.
- code-review-graph - Graphe d'intelligence locale du code pour MCP et CLI ; construit une carte persistante du codebase pour les outils IA.
- kimi-code - Le CLI agent Kimi Code : point de depart pour les agents de nouvelle generation.
- jcode - Le harnais d'agent le plus intelligent pour le code.
- awesome-llm-apps - Plus de 100 applications IA, competences d'agents et apps RAG - libre et open source.
- pi (2026-09-08) - Boite a outils agent IA : API LLM unifiee, boucle agent, TUI, CLI agent de codage.
- openinterpreter - Agent de codage pour les modeles ouverts comme Kimi K3.
- i-have-adhd - Skill pour agent de codage qui arrete le discour de cacher la reponse. Sortie ADHD-friendly.
- pi-web - Interface web pour agent de codage PI avec commandes et interfaces.
- skills - Skills provenant directement du repertoire .agents d'un ingenieur reelement experimente.
- text-to-cad - Collection de skills agent pour CAO, robotique et conception hardware.
- book-to-skill - Transforme tout PDF technique en competence Claude Code : pret a etudier, referencer et utiliser pendant le travail.. (13726 stars)
- adhd - Skill agent de codage : tree-of-thought avec pruning, base sur Claude & Codex Agent SDK. Ideation divergente.. (2879 stars)
- openwork - Alternative open-source a Claude Cowork (base sur opencode) : harness d agents de codage auto-heberge, multi-agents et multi-modeles.* ego-lite - Navigateur optimise pour les agents IA : automation avec etat de session, compatible Codex et Claude Code.. (6536 stars)
- hermes-agent (2026-09-06) : The agent that grows with you (242,475⭐)
- humanizer (2026-09-06) : Agent skill that removes signs of AI-generated writing from text (44,127⭐)
- netdata (2026-09-06) : The fastest path to AI-powered full stack observability, even for lean teams. (80,444⭐)
- openclaw (2026-09-06) : The AI that really does things. Any OS. Any Platform. The lobster way. 🦞 (389,039⭐)
- ClickHouse (2026-09-06) : ClickHouse® is a real-time analytics database management system (49,671⭐)
- impeccable (2026-09-06) : The design language that makes your AI harness better at design. (66,038⭐)
- cherry-studio (2026-09-08) : AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs (51,518⭐)
- n8n (2026-09-06) : Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations. (203,547⭐)
- LibreChat (2026-09-06) : Enhanced ChatGPT Clone: Features Agents, MCP, Skills, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, Mistral, OpenRouter, Vertex AI, Gemini, Artifacts, AI model switching, message search, Code Interpreter, langchain, DALL-E-3, OpenAPI Actions, Functions, Secure Multi-User Auth, Presets, open-source for self-hosting. Active (42,869⭐)
- Summer2027-Internships (2026-09-06) : Summer 2027 software engineering, data science, AI, quant, product management, and hardware internship postings. Updated daily by Simplify and Pitt CSC. (47,174⭐)
- ai-engineering-from-scratch (2026-09-06) : Learn it. Build it. Ship it for others. (52,577⭐)
- CopilotKit (2026-09-06) : The Frontend Stack for Agents & Generative UI. React, Angular, Mobile, Slack, and more. Makers of the AG-UI Protocol (37,215⭐)
- litellm (2026-09-06) : The fastest, litest AI Gateway. Rust core with Python SDK. Call 100+ LLM APIs in OpenAI (or native) format with cost tracking, guardrails, load balancing, and logging [Bedrock, Azure, OpenAI, Anthropic, OpenAI, VertexAI, vLLM, Nvidia NIM] (58,161⭐)
- AutoGPT (2026-09-06) : AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters. (187,174⭐)
- agno (2026-09-06) : Build, run, and manage agent platforms. (42,070⭐)
- reactive-resume (2026-09-06) : A one-of-a-kind resume builder that keeps your privacy in mind. Completely secure, customizable, portable, open-source and free forever. Try it out today! (42,254⭐)
- langchain (2026-09-06) : The agent engineering platform. (145,805⭐)
- frigate (2026-09-06) : NVR with realtime local object detection for IP cameras (35,671⭐)
- open-webui (2026-09-06) : User-friendly AI Interface (Supports Ollama, OpenAI API, ...) (151,125⭐)
- sglang (2026-09-06) : SGLang is a high-performance serving framework for large language models and multimodal models. (35,535⭐)
- worldmonitor (2026-09-06) : Real-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface (85,699⭐)
- lobehub (2026-09-08) : 🤯 LobeHub is your Chief Agent Operator, organizing your agents into 7×24 operations by hiring, scheduling, and reporting on your entire AI team. (82,273⭐)
- DeepSeek-Reasonix (2026-09-06) : DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running. (35,427⭐)
- last30days-skill (2026-09-06) : AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary (61,434⭐)
- airflow (2026-09-06) : Apache Airflow - A platform to programmatically author, schedule, and monitor workflows (46,754⭐)
- llama.cpp (2026-09-06) : LLM inference in C/C++ (127,269⭐)
- vllm (2026-09-06) : A high-throughput and memory-efficient inference and serving engine for LLMs (91,095⭐)
- OpenHands (2026-09-06) : 🙌 OpenHands: AI-Driven Development (86,361⭐)
- ai-job-search (2026-09-06) : The job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it. (41,028⭐)
- transformers (2026-09-06) : 🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training. (164,905⭐)
- oh-my-openagent (2026-09-06) : OmO: Drop your tokens. Ultrawork. Done. (68,760⭐)
- awesome-claude-code (2026-09-06) : A hand-picked collection of the finest of resources for the most awesome of agents, Claude Code, the undisputed champion of coding companions, from the unstoppable team at Anthropic PBC. A delectable showcase of top tier skills, ambidextrous agents, scintillating status lines, top notch developer tooling, and also we have plugins (53,598⭐)
- strix (2026-09-06) : Open-source AI penetration testing tool to find and fix your app’s vulnerabilities. (60,917⭐)
- posthog (2026-09-06) : :hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP. (39,613⭐)
- dbeaver (2026-09-06) : Free universal database tool and SQL client (51,675⭐)
- siyuan (2026-09-06) : An open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together 开源、隐私优先、自托管的知识工作空间,让人与智能体在此协作 (46,201⭐)
- AstrBot (2026-09-06) : AI Agent Assistant & development framework that integrates lots of IM platforms, LLMs, plugins and AI feature, and can be your openclaw alternative. ✨ (40,094⭐)
- PDFMathTranslate (2026-09-06) : [EMNLP 2025 Demo] PDF scientific paper translation with preserved formats - 基于 AI 完整保留排版的 PDF 文档全文双语翻译,支持 Google/DeepL/Ollama/OpenAI 等服务,提供 CLI/GUI/MCP/Docker/Zotero (36,738⭐)
- nanobot (2026-09-06) : Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps (47,763⭐)
- airi (2026-09-06) : 💖🧸 Self hosted, you-owned Grok Companion, a container of souls of waifu, cyber livings to bring them into our worlds, wishing to achieve Neuro-sama's altitude. Capable of realtime voice chat, Minecraft, Factorio playing. Web / macOS / Windows supported. (48,836⭐)
- langextract (2026-09-06) : A Python library for extracting structured information from unstructured text using LLMs with precise source grounding and interactive visualization. (38,547⭐)
- tidb (2026-09-06) : TiDB is built for agentic workloads that grow unpredictably, with ACID guarantees and native support for transactions, analytics, and vector search. No data silos. No noisy neighbors. No infrastructure ceiling. (40,493⭐)
- twenty (2026-09-06) : The open alternative to Salesforce, designed for AI. (56,354⭐)
- firecrawl (2026-09-08) : The context API to search, scrape, and interact with the web at scale. 🔥 (177,245⭐)
- agency-agents (2026-09-06) : A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables. (150,446⭐)
- awesome-copilot (2026-09-06) : Community-contributed instructions, agents, skills, and configurations to help you make the most of GitHub Copilot. (38,688⭐)
- hyperframes (2026-09-08) : Write HTML. Render video. Built for agents. (44,348⭐)
- langfuse (2026-09-07) : 🪢 Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datasets. Integrates with OpenTelemetry, LangChain, OpenAI SDK, LiteLLM, and more. 🍊YC W23 (34,261⭐)
- OmniRoute (2026-09-07) : Never stop coding. Free MIT AI gateway: one endpoint, 352 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 550+ contributors (62,013⭐)
- qdrant (2026-09-07) : Qdrant - High-performance, massive-scale Vector Database and Vector Search Engine for the next generation of AI. Also available in the cloud https://cloud.qdrant.io/ (34,411⭐)
- browser (2026-09-07) : Lightpanda: the headless browser designed for AI and automation (34,596⭐)
- paperless-ngx (2026-09-08) : A community-supported supercharged document management system: scan, index and archive all your documents (44,898⭐)
- milvus (2026-09-07) : Milvus is a high-performance, cloud-native vector database built for scalable vector ANN search (46,000⭐)
- rtk (2026-09-07) : CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies (79,117⭐)
- caveman (2026-09-07) : 🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman (103,934⭐)
- supabase (2026-09-07) : The Postgres development platform. Supabase gives you a dedicated Postgres database to build your web, mobile, and AI applications. (108,912⭐)
- awesome-mcp-servers (2026-09-07) : A collection of MCP servers. (94,484⭐)
- llmfit (2026-09-07) : Hundreds of models & providers. One command to find what runs on your hardware. (34,972⭐)
- career-ops (2026-09-07) : Open-source AI job search: scan job portals, evaluate listings into a structured A-H report with a global 1-5 score, tailor your CV, track applications — runs locally in your AI coding CLI (Claude Code, Codex, OpenCode, Antigravity…) (70,337⭐)
- nacos (2026-09-07) : an easy-to-use dynamic service discovery, configuration and service management platform for building AI cloud native applications. (33,349⭐)
- OpenSpec (2026-09-07) : Spec-driven development (SDD) for AI coding assistants. (67,458⭐)
- gemini-cli (2026-09-07) : An open-source AI agent that brings the power of Gemini directly into your terminal. (106,837⭐)
- claude-mem (2026-09-07) : Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More (93,353⭐)
- ToolJet (2026-09-07) : Open-source foundation of ToolJet AI - the enterprise app generation platform for internal tools, dashboards, business applications, workflows and AI agents. Build visually, from a prompt, or from Claude Code, Codex and Cursor over MCP 🚀 (40,858⭐)
- ai-agent-book (2026-09-07) : 《深入理解 AI Agent:设计原理与工程实践》(李博杰 著)开源主仓库:全书正文、编译版 PDF 与按章配套代码 (44,976⭐)
- MinerU (2026-09-07) : Transforms complex documents like PDFs and Office docs into LLM-ready markdown/JSON for your Agentic workflows. (79,343⭐)
- QwenPaw (2026-09-07) : Your Personal AI Assistant; easy to install, deploy on your own machine or on the cloud; supports multiple chat apps with easily extensible capabilities. (34,973⭐)
- goose (2026-09-07) : an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM (53,977⭐)
- ray (2026-09-07) : Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads. (43,721⭐)
- claude-code-best-practice (2026-09-07) : from vibe coding to agentic engineering - practice makes claude perfect (65,694⭐)
- 1Panel (2026-09-07) : 🔥 1Panel is a modern, open-source Linux server management panel and a lightweight AI management platform. (36,820⭐)
- ppt-master (2026-09-07) : AI turns documents or topics into real, native PowerPoint decks—with native shapes, transitions and animations, data-backed charts and tables on demand, audio narration from speaker notes, and support for your own .pptx templates. · by Hugo He (52,530⭐)
- MoneyPrinterTurbo (2026-09-07) : 利用 AI 大模型和自动化工作流,根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow. (121,195⭐)
- multica (2026-09-07) : Make humans and AI agents work as one team — open-source and self-hostable. (49,076⭐)
- ECC (2026-09-08) : The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond. (251,695⭐)
- agent-browser (2026-09-07) : Browser automation CLI for AI agents (42,067⭐)
- new-api (2026-09-07) : A unified AI model hub for aggregation & distribution. It supports cross-converting various LLMs into OpenAI-compatible, Claude-compatible, or Gemini-compatible formats. A centralized gateway for personal and enterprise model management. (47,463⭐)
- LocalAI (2026-09-07) : LocalAI is the open-source AI engine. Run any model - LLMs, vision, voice, image, video - on any hardware. No GPU required. (48,945⭐)
- jan (2026-09-07) : Jan is an open source alternative to ChatGPT that runs 100% offline on your computer. (44,367⭐)
- crewAI (2026-09-07) : Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks. (58,177⭐)
- docling (2026-09-07) : Get your documents ready for gen AI (66,085⭐)
- airllm (2026-09-07) : AirLLM 70B inference with single 4GB GPU (33,802⭐)
- BMAD-METHOD (2026-09-07) : Breakthrough Method for Agile Ai Driven Development (52,743⭐)
- openhuman (2026-09-07) : OpenHuman is an open source personal AI for Mac, Windows and Linux — local-first memory, agent orchestration, and deep research. (39,487⭐)
- context7 (2026-09-07) : Context7 Platform -- Up-to-date code documentation for LLMs and AI code editors (61,719⭐)
- odysseus (2026-09-07) : Self-hosted AI workspace. (86,942⭐)
- meilisearch (2026-09-07) : A lightning-fast search engine API bringing AI-powered hybrid search to your sites and applications. (59,205⭐)
- scientific-agent-skills (2026-09-07) : Turn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 190,000+ scientists worldwide. 165 ready-to-use validated skills plus 100+ scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard. (43,473⭐)
- continue (2026-09-07) : open-source coding agent (35,821⭐)
- crawl4ai (2026-09-07) : 🚀🤖 Crawl4AI: Open-source LLM Friendly Web Crawler & Scraper. Don't be shy, join here: https://discord.gg/jP8KfhDhyN (81,859⭐)
- CowAgent (2026-09-07) : Open-source super AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-model, multi-channel. Lightweight, extensible, one-line install. (formerly chatgpt-on-wechat) (46,805⭐)
- dokploy (2026-09-07) : Open Source Alternative to Vercel, Netlify and Heroku. (37,148⭐)
- system_prompts_leaks (2026-09-07) : Extracted system prompts from Anthropic - Claude Fable 5.1, Opus 5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro, Antigravity. xAI - Grok, Grok Bot, Cursor, Kimi and more! Updated regularly. (64,319⭐)
- Deep-Live-Cam (2026-09-07) : real time face swap and one-click video deepfake with only a single image (96,529⭐)
- browser-use (2026-09-07) : 🌐 Make websites accessible for AI agents. Automate tasks online with ease. (112,929⭐)
- ollama (2026-09-07) : Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. (180,407⭐)
- Fabric (2026-09-07) : Fabric is an open-source framework for augmenting humans using AI. It provides a modular system for solving specific problems using a crowdsourced set of AI prompts that can be used anywhere. (43,866⭐)
- TradingAgents (2026-09-07) : TradingAgents: Multi-Agents LLM Financial Trading Framework (102,888⭐)
- anything-llm (2026-09-08) : Stop renting your intelligence. Own it with AnythingLLM. Everything you need for a powerful local-first agent experience (65,745⭐)
- mempalace (2026-09-08) : The best-benchmarked open-source AI memory system. And it's free. (58,904⭐)
- netron (2026-09-08) : Visualizer for neural network, deep learning and machine learning models (33,447⭐)
- LlamaFactory (2026-09-08) : Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024) (74,639⭐)
- mem0 (2026-09-08) : The Memory Layer for AI Agents - Drop-in memory infrastructure for AI agents and apps. Context that persists. Built for production. (64,892⭐)
- shannon (2026-09-08) : Shannon is an AI pentester for web applications and APIs. It analyzes your source code, identifies attack vectors, and executes real exploits to prove vulnerabilities before they reach production. (47,843⭐)
- google-research (2026-09-08) : Google Research (38,712⭐)
- project-nomad (2026-09-08) : Project NOMAD is an offline-first knowledge and education server. Wikipedia, thousands of books, courses, maps, and optional local AI, all running on hardware you own with no internet required. (36,588⭐)
- JeecgBoot (2026-09-08) : 【低代码v2.0,一句话即可生成整个系统】企业级AI低代码平台,一键生成前后端代码甚至整个系统。 AI Skills 一句话画流程、设计表单、生成报表、大屏。内置 AI应用平台涵盖:AI聊天、知识库、流程编排、MCP插件等,兼容主流大模型。引领AI低代码「Skills 生成 → 在线配置 → 代码生成 → 手工合并->AI修改」开发模式,解决 Java 项目 90% 重复工作,提高效率又不失灵活。 (47,673⭐)
- spec-kit (2026-09-08) : 💫 Toolkit to help you get started with Spec-Driven Development (134,058⭐)
- langgraph (2026-09-08) : Build resilient agents. (41,263⭐)
- next-ai-draw-io (2026-09-08) : A next.js web application that integrates AI capabilities with draw.io diagrams. This app allows you to create, modify, and enhance diagrams through natural language commands and AI-assisted visualization. (35,706⭐)
- superpowers (2026-09-08) : An agentic skills framework & software development methodology that works. (283,246⭐)
Audio & TTS
- [https://huggingface.co/Supertone/supertonic-3 Superto
- VibeVoice (Microsoft) - Framework vocal open-source (TTS + ASR) de Microsoft : synthese, clonage et reconnaissance vocale de pointe.nic-3] — TTS léger pour inférence locale, ONNX Runtime, zéro cloud
- faster-whisper (mutualisé) — Transcription speech-to-text optimisée
- hyperframes (2026-09-08) : Write HTML. Render video. Built for agents. Génération vidéo par agents (44,346⭐)
- Qwen3-Omni-30B-A3B-Instruct — Modèle multimodal Qwen (audio + texte + image)
- nemotron-3.5-asr-streaming-0.6b — Modèle ASR streaming NVIDIA, faible latence pour transcription temps réel
Génération & Édition d'Images
- HiDream-O1-Image — Modèle unifié pixel-level (UiT), sans VAE externe — t2i, édition, personnalisation jusqu'à 2048×2048
- ComfyUI (2026-09-08) : The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. (131,774⭐)
- unsloth (2026-09-06) : Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4, MiniMax-H3, Gemma 4, FLUX and more. (75,722⭐)
RAG & Traitement de Documents
- firecrawl (2026-09-08) : The context API to search, scrape, and interact with the web at scale. 🔥 (177,225⭐)
- RAG sur PDF avec images
- granite-docling-258M — Parsing structuré de documents IBM Granite
- Haystack — Framework RAG end-to-end (deepset)
- paperless-ngx (2026-09-08) : Document management: scan, index, archive. ML, OCR, PDF (45,407⭐)
- graphify (2026-09-08) : Transformer tout codebase (docs, SQL, configs, PDFs) en graphe de connaissances interrogeable (115,000⭐)
- Mem0 — Mémorie à long terme pour agents IA
- meilisearch — Moteur de recherche full-text
- Stirling-PDF - Application PDF #1 sur GitHub : traitement complet des PDF sur tous les appareils.
- pdf-inspector — Fast Rust library for PDF inspection, classification, and text extraction. Detects scanned vs text-based PDFs for smart routing.
- Ossie - Standard open-source Apache pour les metadonnees semantiques entre analytics, IA et BI.
- LightRAG (2026-09-06) : [EMNLP2025] LightRAG: Simple and Fast Retrieval-Augmented Generation (39,438⭐)
- appwrite (2026-09-06) : Appwrite® - complete cloud infrastructure for your web, mobile and AI apps. Including Auth, Databases, Storage, Functions, Messaging, Hosting, Realtime and more (57,303⭐)
- dify (2026-09-07) : Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack. (154,647⭐)
- PageIndex (2026-09-07) : 📑 PageIndex: Document Index for Vectorless, Reasoning-based RAG (35,552⭐)
- graphrag (2026-09-07) : A modular graph-based Retrieval-Augmented Generation (RAG) system (35,859⭐)
- ragflow (2026-09-07) : RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs (90,149⭐)
- headroom (2026-09-07) : Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server. (69,152⭐)
- OpenViking (2026-09-07) : Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills. (35,812⭐)
- pathway (2026-09-07) : Python ETL framework for stream processing, real-time analytics, LLM pipelines, and RAG. (62,335⭐)
- ruflo (2026-09-07) : 🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated (71,110⭐)
- deer-flow (2026-09-07) : An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours. (81,658⭐)
- awesome-llm-apps (2026-09-07) : 100+ AI Agents, Agent Skills and RAG Apps - Free and Open Source. (136,453⭐)
- private-gpt (2026-09-07) : Complete API layer for private AI applications on local models: RAG, skills, tools, MCP, text-to-sql, and more. Works with any OpenAI-compatible inference server. (57,498⭐)
APIs à Développer
- Classificateur IA — Classification de contenu
- Résumé mutualisé — API de résumé de texte partagée
- NER — Reconnaissance d'entités nommées
- Compressor — Compression de contenu
Infrastructure & Backend
- OmniRoute — Passerelle IA gratuite : un endpoint, 231+ providers (50+ free), compression 15-95% tokens.
- Temporal — Orchestration de workflows critiques
- Semantic Router — Routage sémantique de requêtes vLLM
- Supabase — Alternative open-source Firebase (PostgreSQL, Auth, etc.)
- Metabase — Analytics et dashboards open-source
- N8N — Workflow automation open-source
- iroh - Stack reseau modulaire en Rust -- decentralise, base sur des cles plutot que des IP.
- LMCache - Couche KV cache ultra-rapide pour les L
- GeoLibre - Plateforme SIG open-source cloud-native pour visualisation et analyse de donnees geospatiales (web, bureau, mobile, Jupyter).
- Instatic - CMS visuel auto-heberge et agentic alternative a Webflow/Framer : utilisateurs, roles, plugins, base de donnees integree.LM -- reduit drastiquement la latence d'inference.
- Meshery - Gestionnaire cloud native open-source -- observabilite et orchestration multi-cluster.
- worldmonitor - Tableau de bord de renseignement global en temps reel : aggregation IA de news et suivi d'infrastructure.
- openship - Plateforme de deploiement auto-hebergeee et auto-deployee avec travaux automatiques.
Outils Dev
- Aider — Assistant de codage IA en ligne de commande
- Continue — Extension IDE IA (VS Code, JetBrains)
- wigolo - Web local-first pour agent de codage — recherche, fetch, crawl et recherche via MCP, sans clee API.
- MCP LLM — Modèle de langage via Model Context Protocol
Assistants IA & outils cloud
Assistants IA
- Assistants IA auto-hébergés
- Hermes WebUI + Ollama + GPU
- Stack typique pour assistant privé, API OpenAI-compatible et expérimentation locale.
Développement, modèles & veille
- Découverte de modèles
- Évaluation & benchmarks
- Outils de développement & fine-tuning
Matériel IA & GPU
- NVIDIA GH200
- DGX Spark
- GROQ LLM accelerator
API Realtime AI (DEV)
Statut : environnement DEV, remplaçante prévue de l’API OpenAI pour les cas temps réel.
Configuration
| Variable | Valeur |
|---|---|
| OPENAI_API_BASE | wss://api-realtime-ai.ailab.infocepo.com:wait-2026-12/v1
|
| OPENAI_API_KEY | sk-XXXXX
|
Dépôt GitHub
Page de test
external-test/half-duplex.html— annulation d’écho + mode half-duplex.
Compatibilité
Remplacer l’URL OpenAI par $OPENAI_API_BASE pour tester compatibilité et performances.
API LLM (OpenAI compatible)
- URL de base :
https://api-nothink.ailab.infocepo.com/v1 - Création du token : OPENAI_API_KEY
- Documentation : Documentation API
Liste des modèles
curl -X GET \ 'https://api-nothink.ailab.infocepo.com/v1/models' \ -H 'Authorization: Bearer sk-XXXXX' \ -H 'accept: application/json' \ | jq | sed -rn 's#^.*id.*: "(.*)".*$#* \1#p' | sort -u
Modèles ouverts & endpoints internes
Dernière mise à jour : 2026-06-30
Les modèles ci-dessous correspondent à des endpoints logiques exposés derrière une passerelle.
| Endpoint | Description / usage principal |
|---|---|
| ai-thinking | qwen3.6 fp8 – thinking |
| ai-fast | qwen3.6 fp8 en mode fast – vision/OCR/ai-default |
| ai-embedding | bge-m3 – recherche sémantique |
| ai-stt | whisper3-turbo – transcription vocale multilingual |
| ai-tts | OmniVoice – TTS multilingual |
| ai-image | OpenDalle – image génération |
Exemple bash
export OPENAI_API_MODEL="ai-default"
export OPENAI_API_BASE="https://api-nothink.ailab.infocepo.com/v1"
export OPENAI_API_KEY="sk-XXXXX"
promptValue="Quel est ton nom ?"
jsonValue='{
"model": "'${OPENAI_API_MODEL}'",
"messages": [{"role": "user", "content": "'${promptValue}'"}],
"temperature": 0
}'
curl ${OPENAI_API_BASE}/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d "${jsonValue}" 2>/dev/null | jq '.choices[0].message.content'
Vue infra LLM
DEV (au choix)
- A.
LiteLLM → vLLM/SgLang: tests perf / compatibilité - B.
LiteLLM → Ollama: simple, rapide à itérer - C.
Ollamadirect : POC ultra-léger
DEV – modèle FR / résumé
LiteLLM → Ollama /v1
PROD
- Standard :
LiteLLM → vLLM/SgLang - Pont DEV→PROD :
LiteLLM (DEV) → LiteLLM (PROD) → vLLM/SgLang
Notes :
- LiteLLM = passerelle unique (clés, quotas, logs)
- vLLM/SgLang = performance / stabilité en charge
- Ollama = simplicité de prototypage
API Image to Text
- Utilise l’API LLM avec un endpoint adapté à l’OCR / VLM.
- Modèle recommandé :
ai-vision
Exemple bash
OPENAI_API_KEY=sk-XXXXX
base64 -w0 "/path/to/image.png" > img.b64
jq -n --rawfile img img.b64 \
'{
model: "ai-vision",
messages: [
{
role: "user",
content: [
{ "type": "text", "text": "Décris cette image." },
{
"type": "image_url",
"image_url": { "url": ("data:image/png;base64," + ($img | rtrimstr("\n"))) }
}
]
}
]
}' > payload.json
curl https://api-nothink.ailab.infocepo.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
--data-binary @payload.json
Exemple Python
import base64
import json
import requests
import os
API_KEY = os.getenv("OPENAI_API_KEY")
MODEL = "ai-vision"
IMG_PATH = "/path/to/image.png"
API_URL = "https://api-nothink.ailab.infocepo.com/v1/chat/completions"
with open(IMG_PATH, "rb") as f:
img_b64 = base64.b64encode(f.read()).decode("utf-8")
payload = {
"model": MODEL,
"messages": [
{
"role": "user",
"content": [
{"type": "text", "text": "Décris cette image."},
{
"type": "image_url",
"image_url": {"url": f"data:image/png;base64,{img_b64}"}
}
]
}
]
}
headers = {
"Authorization": f"Bearer {API_KEY}",
"Content-Type": "application/json"
}
response = requests.post(API_URL, headers=headers, data=json.dumps(payload))
if response.ok:
print(json.dumps(response.json(), indent=2, ensure_ascii=False))
else:
print(f"Erreur {response.status_code}: {response.text}")
API STT
- URL :
https://api-audio2txt.ailab.infocepo.com/v1 - Clé :
OPENAI_API_KEY=sk-XXXXX - Modèle :
whisper-1 - Documentation : API STT docs
Exemple Python
import requests
OPENAI_API_KEY = 'sk-XXXXX'
url = 'https://api-audio2txt.ailab.infocepo.com/v1/audio/transcriptions'
headers = {
'Authorization': f'Bearer {OPENAI_API_KEY}',
}
files = {
'file': ('file.opus', open('/path/to/file.opus', 'rb')),
'model': (None, 'whisper-1')
}
response = requests.post(url, headers=headers, files=files)
print(response.json())
Exemple curl
[ ! -f /tmp/test.ogg ] && wget "https://upload.wikimedia.org/wikipedia/commons/1/17/Fables_de_La_Fontaine_Livre_1_01.ogg" -O /tmp/test.ogg export OPENAI_API_KEY=sk-XXXXX curl https://api-audio2txt.ailab.infocepo.com/v1/audio/transcriptions \ -H "Authorization: Bearer $OPENAI_API_KEY" \ -F model="whisper-1" \ -F file="@/tmp/test.ogg"
Notes
- Plusieurs formats audio sont acceptés.
- Le flux final est normalisé en 16 kHz mono.
- Pour une qualité optimale : privilégier OPUS 16 kHz mono.
UI
API TTS
- URL :
https://api-tts-omnivoice.ailab.infocepo.com/v1 - Clé :
OPENAI_API_KEY=sk-XXXXX - Documentation : API TTS docs
Exemple
export OPENAI_API_KEY=sk-XXXXX
curl https://api-tts-omnivoice.ailab.infocepo.com/v1/audio/speech \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4o-mini-tts",
"input": "Bonjour, ceci est un test de synthèse vocale.",
"voice": "coral",
"instructions": "Speak in a cheerful and positive tone.",
"response_format": "opus"
}' | ffplay -i -
API Text to Image
- URL :
https://api-txt2image.ailab.infocepo.com/v1 - Clé API :
OPENAI_API_KEY=sk-... - Documentation : API TXT2IMAGE docs
Exemple
export OPENAI_API_KEY=EMPTY
curl https://api-txt2image.ailab.infocepo.com/v1/images/generations \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"prompt": "a photo of a happy corgi puppy sitting and facing forward, studio light, longshot",
"n": 1,
"size": "1024x1024"
}'
API Diarization
- Documentation : API Diarization docs
Exemple
wget "https://upload.wikimedia.org/wikipedia/commons/6/60/Mike_Peters_on_Politics_and_Emotion_%28Interview_1984%29.mp3" -O /tmp/test.mp3 curl -X POST "https://api-diarization.ailab.infocepo.com/upload-audio/" \ -H "Authorization: Bearer token1" \ -F "file=@/tmp/test.mp3"
API Summary
- Documentation : API Summary docs
Exemple
text="The tower is 324 metres tall and is one of the most recognizable monuments in the world."
json_payload=$(jq -nc --arg text "$text" '{"text": $text}')
curl -X POST https://api-summary.ailab.infocepo.com:wait-2026-12/summary/ \
-H "Content-Type: application/json" \
-d "$json_payload"
API Text Embeddings
- URL :
https://api-embedding.ailab.infocepo.com/v1 - Documentation : Documentation
Exemple
curl https://api-embedding.ailab.infocepo.com/v1/embeddings \
-X POST \
-d '{"model":"bge-m3","input":"What is Deep Learning?"}' \
-H 'Content-Type: application/json'
API DB Vectors (ChromaDB)
Production
- URL :
https://chromadb.ailab.infocepo.com:wait-2026-12 - Token :
XXXXX
Lab
export CHROMA_HOST=https://chromadb.c1.ailab.infocepo.com:wait-2026-12 export CHROMA_PORT=443 export CHROMA_TOKEN=XXXX
Exemple curl
curl -v "${CHROMA_HOST}"/api/v1/collections \
-H "Authorization: Bearer ${CHROMA_TOKEN}"
Exemple Python
import chromadb
from chromadb.config import Settings
def chroma_http(host, port=80, token=None):
return chromadb.HttpClient(
host=host,
port=port,
ssl=host.startswith('https') or port == 443,
settings=(
Settings(
chroma_client_auth_provider='chromadb.auth.token.TokenAuthClientProvider',
chroma_client_auth_credentials=token,
) if token else Settings()
)
)
client = chroma_http(CHROMA_HOST, CHROMA_PORT, CHROMA_TOKEN)
collections = client.list_collections()
print(collections)
Déployer sa propre instance
export nameSpace=your_namespace
domainRoot=ailab.infocepo.com
helm repo add chroma https://amikos-tech.github.io/chromadb-chart/
helm repo update
helm upgrade --install chromadb chroma/chromadb -n ${nameSpace} \
--set chromadb.apiVersion="0.4.24" \
--set ingress.enabled=true \
--set ingress.hosts[0].host="${nameSpace}-chromadb.${domainRoot}" \
--set ingress.hosts[0].paths[0].path=/ \
--set ingress.hosts[0].paths[0].pathType=ImplementationSpecific \
--set ingress.annotations."cert-manager\.io/cluster-issuer"=letsencrypt-prod \
--set ingress.tls[0].secretName=${nameSpace}-chromadb.${domainRoot}-tls \
--set ingress.tls[0].hosts[0]="${nameSpace}-chromadb.${domainRoot}"
kubectl -n ${nameSpace} patch ingress/chromadb --type=json \
-p '[{"op":"add","path":"/metadata/annotations/nginx.ingress.kubernetes.io~1proxy-body-size","value":"0"}]'
Récupérer le token
kubectl --namespace ${nameSpace} get secret chromadb-auth \
-o jsonpath="{.data.token}" | base64 --decode && echo
Registry
- URL : registry.ailab.infocepo.com:wait-2026-09
- Login :
user - Password :
XXXXX
Exemple
curl -u "user:XXXXX" https://registry.ailab.infocepo.com:wait-2026-09/v2/_catalog
Exemple K8S
deploymentName=
nameSpace=
kubectl -n ${nameSpace} create secret docker-registry pull-secret \
--docker-server=registry.ailab.infocepo.com:wait-2026-09 \
--docker-username=user \
--docker-password=XXXXX \
--docker-email=contact@example.com
kubectl -n ${nameSpace} patch deployment ${deploymentName} \
-p '{"spec":{"template":{"spec":{"imagePullSecrets":[{"name":"pull-secret"}]}}}}'
Stockage objet externe (S3)
- Endpoint :
https://s3.ailab.infocepo.com:wait-2026-09 - Access key :
XXXX - Secret key :
XXXX
Un bucket nommé ORG a été créé pour stocker des documents de démonstration.
RAG optimisation
- Embeddings :
BAAI/bge-m3 chunk_size=1000chunk_overlap=100- LLM :
qwen3.6 - Pour les PDF mixtes : PDF → image → OCR / VLM peut améliorer les résultats.
Workflow
Autoroute directe : code → CI/CD → deploy → monitoring. Pas de comité, pas de responsable désigné — chaque étape se déclenche automatiquement si la précédente passe. Si ça rate, alerte directe sans passer par un humain sauf décision explicite de fallback.
Environnements
Hors production
- Utiliser datalab
- Support : canal Mattermost Offre IA
- Le pseudo utilisateur doit respecter la convention interne
- Demander si besoin un accès Linux + Kubernetes
Production (best-effort)
- Publier le code applicatif, les secrets (format SOPS), le Dockerfile et le code infra (Helm ou manifests K8S) sur Git
- Demander un namespace
- Lire la documentation de surveillance associée
Limites de l’infrastructure
- Les charges GPU sont intentionnellement limitées en journée.
- Coût unitaire mesuré automatiquement (dashboards temps réel, pas Excel).
Cloud Lab & projets d’audit
Le Cloud Lab fournit des scénarios reproductibles : audit d’infrastructure, migration cloud, automatisation, haute disponibilité.
Projet d’audit
Script Bash d’audit permettant de :
- détecter les dérives de configuration,
- comparer plusieurs environnements,
- préparer un plan de migration ou de remédiation.
- open-code-review - Outil de review de code open-source teste a l echelle Alibaba : pipeline deterministe + agent LLM, precise par ligne.. (16590 stars)
Exemple de migration cloud
| Tâche | Description | Durée (jours) |
|---|---|---|
| Audit infrastructure | 82 services, audit automatisé via ServerDiff.sh | 1.5 |
| Diagramme d’architecture | Conception visuelle et documentation | 1.5 |
| Contrôles de conformité | 2 clouds, 6 hyperviseurs, 6 To RAM | 1.5 |
| Installation plateforme cloud | Déploiement des environnements cibles | 1.0 |
| Vérification de stabilité | Premiers tests fonctionnels | 0.5 |
| Étude d’automatisation | Identification des tâches répétitives | 1.5 |
| Développement des templates | 6 templates, 8 environnements, 2 clouds / OS | 1.5 |
| Diagramme de migration | Illustration du processus | 1.0 |
| Écriture du code de migration | 138 lignes (voir MigrationApp.sh) | 1.5 |
| Stabilisation | Validation de la reproductibilité | 1.5 |
| Benchmark cloud | Comparaison vs legacy | 1.5 |
| Réglage des temps d’arrêt | Calcul du downtime | 0.5 |
| Chargement VM | 82 VMs : OS, code, 2 IP par VM | 0.1 |
| Total | 15 jours.homme | |
Vérifications de stabilité (HA minimale)
| Action | Résultat attendu |
|---|---|
| Extinction d’un nœud | Tous les services redémarrent automatiquement sur les autres nœuds |
| Extinction / redémarrage simultané de tous les nœuds | Les services repartent correctement après reboot |
Autonomie testée : vérifier que la migration fonctionne seule en tirant le repo et lançant le script, sans personne qui connaît le système de tête.
Architecture web & bonnes pratiques
Principes de conception :
- privilégier une infrastructure simple, modulaire et flexible,
- rapprocher le contenu du client (GDNS ou équivalent),
- utiliser des load balancers réseau (LVS, IPVS),
- comparer les coûts et éviter le vendor lock-in,
- pour TLS :
- HAProxy pour les frontends rapides,
- Envoy pour les cas avancés (mTLS, HTTP/2/3),
- pour le cache :
- Varnish, Apache Traffic Server,
- favoriser les stacks open-source,
- utiliser files, buffers, queues et quotas pour lisser les pics.
Références
Comparatif des grandes plateformes cloud
| Fonctionnalité | Kubernetes | OpenStack | AWS | Bare-metal | HPC | CRM | oVirt |
|---|---|---|---|---|---|---|---|
| Outils de déploiement | Helm, YAML, ArgoCD, Juju | Ansible, Terraform, Juju | CloudFormation, Terraform, Juju | Ansible, Shell | xCAT, Clush | Ansible, Shell | Ansible, Python |
| Méthode de bootstrap | API | API, PXE | API | PXE, IPMI | PXE, IPMI | PXE, IPMI | PXE, API |
| Contrôle routeur | Kube-router | Router/Subnet API | Route Table / Subnet API | Linux, OVS | xCAT | Linux | API |
| Contrôle firewall | Istio, NetworkPolicy | Security Groups API | Security Group API | Linux firewall | Linux firewall | Linux firewall | API |
| Virtualisation réseau | VLAN, VxLAN | VPC | VPC | OVS, Linux | xCAT | Linux | API |
| DNS | CoreDNS | DNS-Nameserver | Route 53 | GDNS | xCAT | Linux | API |
| Load balancer | Kube-proxy, LVS | LVS | Network Load Balancer | LVS | SLURM | Ldirectord | N/A |
| Stockage | Local, cloud, PVC | Swift, Cinder, Nova | S3, EFS, EBS, FSx | Swift, XFS, EXT4, RAID10 | GPFS | SAN | NFS, SAN |
Cette table sert de point de départ pour choisir la bonne stack selon :
- le niveau de contrôle souhaité,
- le contexte (on-prem, cloud public, HPC…),
- les outils d’automatisation existants.
Haute disponibilité, HPC & DevSecOps
Haute disponibilité avec Corosync & Pacemaker
Principes :
- clusters multi-nœuds ou multi-sites,
- fencing via IPMI,
- provisioning PXE / NTP / DNS / TFTP,
- pour 2 nœuds : attention au split-brain,
- 3 nœuds ou plus recommandés en production.
Ressources fréquentes
- multipath, LUNs, LVM, NFS,
- processus applicatifs,
- IP virtuelles, DNS, listeners réseau.
HPC
- orchestration de jobs (SLURM ou équivalent),
- stockage partagé haute performance,
- intégration possible avec des workloads IA.
DevSecOps
- CI/CD avec contrôles de sécurité intégrés,
- observabilité dès la conception,
- scans de vulnérabilité,
- gestion des secrets,
- policy-as-code.
News & trends
Formation & apprentissage
- Transformers Explained
- Labs, scripts et retours d’expérience concrets dans le projet Cloud Lab
Liens cloud & IT utiles
- Cloud Providers Compared
- Global Internet Topology Map
- CNCF Official Landscape
- Wikimedia Cloud Wiki
- OpenAPM
- Red Hat Package Browser
- Baromètre TJM IT
- Indicateurs salariaux IT
Outils collaboratifs
Dépôts de code
Base de connaissance
Messagerie
- contact interne / support selon les projets
SSO
À propos & contributions
Suggestions de corrections, améliorations de schémas, retours d’expérience ou nouveaux labs bienvenus.
Ce wiki a vocation à rester un laboratoire vivant pour l’IA, le cloud et l’automatisation.


