MCP App StoreLe meilleur de l'écosystème IA et MCP, sélectionné chaque jour.
Sources
Introduction of Visual Salamandra, a multimodal model focusing on enhanced visual understanding and reasoning. This development pushes the boundaries of how AI interprets complex visual data alongside text.
Hugging Face welcomes Llama 4 Maverick and Scout, expanding the availability of high-performance open-weights models. This release provides developers with new options for fine-tuning and deploying advanced LLMs.
Hugging Face details its approach to managing secrets at scale for massive AI infrastructure. A practical look at ensuring security without sacrificing developer velocity in high-compute environments.
Text Generation Inference (TGI) now supports Intel Gaudi accelerators. This provides an alternative to NVIDIA GPUs for high-performance LLM serving and inference.
The latest update on the Open R1 project, focusing on the replication of DeepSeek-R1's reasoning capabilities. It provides insights into the training process and current performance of open-source reasoning models.
A deep dive into training and fine-tuning reranker models using the Sentence Transformers library. Essential for developers building high-precision RAG pipelines and semantic search systems.
Gradio introduces an upgraded Dataframe component to improve how AI apps handle and display tabular data. This update streamlines data interaction for developers building internal tools and model dashboards.
Anthropic introduced a dedicated "think" tool that gives Claude a scratchpad to reason through complex multi-step tool use situations before acting. In tests, adding the think tool significantly improved accuracy on tasks requiring careful sequencing of tool calls. A simple but effective technique any agent developer can adopt immediately.
Anthropic achieved then-state-of-the-art performance on SWE-bench Verified with Claude 3.5 Sonnet using only a bash tool and a text editor — no scaffolding tricks. The post details the technical approach and what the benchmark results imply about real-world software engineering capability. A landmark result that established Claude as a serious coding model.
Anthropic's foundational guide to building reliable AI agents — covering when to use agents vs. simpler pipelines, how to design tool interfaces, how to handle failures gracefully, and the core patterns (augmented LLM, routing, parallelisation, orchestrator-subagent) that underpin most production agent systems. Still one of the best starting points for anyone new to agent development.
Anthropic introduced Contextual Retrieval — a technique that prepends chunk-specific context to each piece of text before embedding, dramatically improving RAG retrieval accuracy. Combined with BM25 reranking, it reduced retrieval failure rates by up to 67% compared to standard RAG. A practical technique that any team using RAG with Claude can adopt today.