Le meilleur du dev et de l'IA, scoré chaque jour par un agent. Filtré, résumé, classé. Aucune couleur, aucun bruit — juste la matière.
Bytebot is a self-hosted AI desktop agent that runs inside a containerized Linux desktop, enabling natural-language tasks across apps, files, and web portals with its own virtual desktop. It features a four-component architecture (Desktop, Agent, UI, APIs), supports multiple AI providers, and offers Docker and Helm deployments with practical API examples for task creation and direct desktop control.
tldraw is a React-based SDK for building on an infinite canvas. It includes multiplayer, drawing and diagramming tools, a runtime API, and full extensibility with custom shapes, tools, bindings and UI components, plus AI integrations and DOM embedding. The project ships with starter kits and a quick-start workflow to bootstrap apps and local development.
LibreChat est une plateforme d’IA en auto-hébergement qui unifie les principaux fournisseurs dans une interface privée et respectueuse de la confidentialité. Elle propose des Agents, le Model Context Protocol (MCP), un Code Interpreter, la recherche web intégrée, la génération et l’édition d’images, ainsi qu’une gestion multilingue et une authentification multi-utilisateurs. Open-source et orientée contrôle d’infrastructure, elle offre de multiples intégrations (OpenAI, Anthropic, AWS Bedrock, V
Chat2API est une plateforme multi-provider qui expose une API OpenAI-compatible pour accéder gratuitement à des modèles IA via leurs UIs officielles. Elle prend en charge DeepSeek, GLM, Kimi, MiniMax, Qwen et Z.ai et s’intègre à des outils comme openlcaw, Cline et Roo-Code, tout en gérant le contexte, les appels de fonction et le suivi en temps réel. L’architecture repose sur Electron/React avec proxy multi-provider, gestion des clés et journaux détaillés.
Open Lovable permet de cloner et recréer rapidement n'importe quel site en une appli React, via une interface IA pour construire des apps. Le repo propose un flux setup avec clone, installation, configuration d'API keys et choix d'un sandbox (Vercel ou E2B), puis un mode dev local.
Pascal Editor is a Turborepo-based 3D building editor built with React Three Fiber and WebGPU. It splits core schemas/state, a viewer for rendering, and editor tooling, using Zustand stores, a node/scene registry, and systems to generate geometry (walls, slabs, roofs) with undo/redo support. The architecture emphasizes a registry and dirty-node processing for efficient re-renders.
Novu is an open-source notification infrastructure with a unified API for delivering across Inbox/In‑App, Email, SMS, Push, and Chat. It offers an embeddable Inbox, a Digest Engine, a No‑Code Email editor, and a Workflow Engine to tailor multi‑channel notifications, plus a single API for all providers. The project follows an open‑core model (MIT core, enterprise features) and targets developers building scalable, multi‑channel notification experiences.
Claude Code Router lets you route Claude Code requests across multiple model providers, with transformers to adapt payloads and on-the-fly model switching via /model. It offers CLI, UI, presets and GitHub Actions integration, plus a plugin system and a configurable router (including a custom router) for advanced routing and cost control.
ClaudePrism est un workspace scientifique hors-ligne combinant LaTeX, Python et 100+ compétences, tournant localement sur desktop via Tauri et Rust. Il propose un assistant Claude intégré, une compilation LaTeX hors ligne avec Tectonic, un gestionnaire Python uv/venv, et un historique Git local avec diffs et restaurations faciles.
DeepSeek-V3 is a 671B Mixture-of-Experts LLM with 37B activated per token, featuring MLA and an auxiliary-load-free load balancing strategy, trained on 14.8T tokens. It achieves top open-source benchmarks and supports FP8-enabled backends (SGLang, LMDeploy, TRT-LLM, vLLM, LightLLM) with local run steps and a Multi-Token Prediction objective plus post-training distillation from DeepSeek-R1.
Translate Books with LLMs permet de traduire des livres et documents localement ou dans le cloud en préservant la mise en forme (EPUB, SRT, DOCX, TXT) et en reprenant là où on s'était arrêté. Le projet propose de multiples fournisseurs (Ollama, Poe, OpenRouter, OpenAI, Gemini, etc.), un CLI complet et des options Docker/ENV pour une intégration rapide.
Sherlock is a CLI tool that searches for a username across 400+ social networks and stores results per site. It supports multiple installation methods (pipx, Docker, or distro packages) and a wide range of options (output formats, proxies, site filtering).
GhostTrack is a CLI OSINT tool to track IPs, phone numbers, and usernames. It provides a menu with IP Tracker, Phone Tracker, and Username Tracker, and can be installed on Linux or Termux and run via Python. It can be extended with the Seeker tool for enhanced IP discovery.
CorridorKey is a neural unmixing-based keyer that, given a raw green-screen frame and a coarse alpha hint, predicts the true foreground color and a linear alpha for every pixel, removing spill and edge artifacts. It offers resolution-invariant inference (2048x2048 backbone with 4K plates), 16/32-bit EXR outputs, and optional AlphaHint generators (GVM and VideoMaMa). The project ships with Docker workflows, a CLI wizard, and multi-backend support (Torch on Linux/Windows and MLX on Apple Silicon),
TradingAgents est un cadre multi-agent LLM pour l'analyse et les décisions de trading, décomposant les tâches en Analystes, Chercheurs, Trader et Gestion des risques. Il prend en charge plusieurs fournisseurs LLM (OpenAI, Gemini, Claude, Ollama) et propose installation, CLI, persistance du journal de décisions et reprise sur erreur, le tout en open source.
Demo of a customer service interface built with the OpenAI Agents SDK, pairing a Python backend for orchestration with a Next.js UI. It showcases agents for triage, flight info, bookings, seats, FAQs, and refunds, plus end-to-end flows (seat changes, cancellations, irregular operations) and guardrails to keep conversations airline-focused.
VibeVoice is an open-source family of frontier voice AI models (ASR and TTS) using 7.5 Hz continuous speech tokenizers and a next-token diffusion framework guided by an LLM. It enables 60-minute long-form ASR with diarization and custom hotwords, and 90-minute multi-speaker TTS with up to 4 speakers and multilingual support. The project ships Colab demos, Hugging Face releases, finetuning code, and a real-time streaming variant.
AI4AnimationPy is a Python framework for AI-driven character animation that unifies motion capture processing, training, inference and rendering in NumPy/PyTorch. It replaces Unity-based pipelines with a modular ECS-style architecture and provides tools for IK, animation modules, real-time rendering, and motion import from GLB/FBX/BVH. A CLI enables batch conversion and it supports Standalone, Headless and Manual execution modes.
OmniRoute is an OpenAI-compatible AI gateway that routes requests across 160+ providers with smart routing, retries and policy-based fallbacks. It bundles MCP and A2A orchestration, a universal /v1 endpoint, and observability dashboards with a free-tier path and a desktop app. It's 100% TypeScript with format translation across providers.
Postiz est un outil open-source de planification automatique des réseaux sociaux, orienté IA et disponible en monorepo. Il expose une CLI agent et s'appuie sur Next.js/NestJS/Prisma/Temporal avec des intégrations (Make, n8n, Zapier) et un SDK NodeJS. Le dépôt expose aussi des informations de conformité et des options de self-hosting et de sponsoring.