Mamba-3
Mamba-3 is introduced as a new open-source State Space Model (SSM) built for inference. It boasts superior performance, being faster than Transformers at decode and stronger than Mamba-2.
Mamba-3 is introduced as a new open-source State Space Model (SSM) built for inference. It boasts superior performance, being faster than Transformers at decode and stronger than Mamba-2.
This workshop offers a deep dive into training AI agents using Reinforcement Learning (RL) principles. It specifically focuses on leveraging open-source tools and techniques for practical agent development.

Este conteúdo anuncia o lançamento dos modelos open-source Qwen2.5-1M (7B e 14B), que agora suportam um comprimento de contexto de até um milhão de tokens. Esta liberação expande as capacidades dos modelos Qwen para lidar com contextos extensos, seguindo a atualização da versão Turbo.
This content explores valuable lessons derived from an analysis of 16 open-source Reinforcement Learning (RL) libraries. It aims to provide insights for practitioners and developers working with RL frameworks.
The content describes porting Anthropic's eval-driven AI skill-creator from Python (for Claude Code) to TypeScript (for OpenCode). The goal is to make this proven methodology for skill creation, testing, and optimization freely accessible to users and over 300 models, removing the paywall of a paid subscription.
Desenvolvedores estão procurando alternativas ao editor de código IA Cursor devido a uma mudança na política de cobrança para um sistema baseado em uso, que gerou custos imprevisíveis para usuários intensivos. Outras razões incluem sua arquitetura fechada e limitações com grandes bases de código, levando à busca por outras opções.
Happy is an open-source CLI wrapper designed to control Claude Code from a phone, solving productivity issues by eliminating waits for permission prompts. This tool, with over 17,000 GitHub stars, is distinct from Anthropic's Claude Cowork Dispatch, which focuses on AI remote control for non-coding tasks.
The content introduces a 2026 guide for running open-source and free LLMs at zero cost, emphasizing practical challenges like rate limits and weak GPU performance experienced when building AI solutions. It highlights the growing importance and accessibility of open-source AI models as a new societal norm.
Hermes Agent is an open-source AI agent by Nous Research designed to run continuously on your server with long-term memory and self-learning capabilities. It can automate tasks like machine management and web app deployment, and connects to mobile-friendly chat platforms.
Tian AI is an open-source, completely offline AI assistant for Android, running a local LLM and a 34GB knowledge base directly on your phone's CPU. It addresses cloud AI privacy concerns by ensuring no data leaves the device, offering a full-stack local AI system without internet or subscription fees.
This article reflects on making Seed4j, an open-source application generator, truly AI-native. It proposes an "AI Triforce" of Power, Wisdom, and Courage to integrate AI agents effectively into its hexagonal architecture.
The content explains the crucial difference between language models that only respond and AI agents capable of searching, reasoning, and remembering. It highlights that agents add tools (like Python functions) to execute code, query databases, and call APIs, overcoming the limitations of LLMs that only generate text.
Foi lançado o sunpeak.ai, um framework de testes autônomo para MCP Apps desenvolvidos para ChatGPT e Claude. Ele replica os ambientes de execução localmente, simplificando drasticamente o complexo processo de teste e suportando testes unitários, E2E e de regressão visual.
An indie developer created a Python application, "2D-level-genie," which uses the Nano Banana API to generate, outpaint, and inpaint 2D game levels, including automatically furnished indoor layers. The project is open-source and aims to assist other indie developers.
Tencent Cloud has open-sourced its Cube Sandbox, a production-grade foundation for deploying AI Agents at industrial scale, addressing challenges like reliability and scalability. The author provides a cautious but optimistic take, seeing it as a much-needed battle-tested platform for engineers.
O protocolo MCP Apps é um padrão de código aberto que define especificações de UI e SDK para chatbots de IA incorporados, buscando padronizar a experiência do usuário e a interoperabilidade. Ele capacita desenvolvedores a integrar assistentes de IA sofisticados em seus produtos.
The author developed an open-source, one-click deployable AI customer service platform to address the repetitive building of infrastructure for bots. This platform integrates multi-channel support, Claude AI, sentiment analysis, and smart escalation, offering a production-ready alternative to costly or complex enterprise solutions.
This article presents the top 10 open-source libraries designed for fine-tuning Large Language Models (LLMs) locally. These tools greatly simplify the fine-tuning process, eliminating the need to build the full training stack from scratch.
This content introduces dotclaude, an MIT-licensed open-source project designed to provide a governance layer for AI-assisted development. It addresses common issues like AI assistants lacking memory of development workflows and the inconsistency this creates across engineering teams.
O autor, frustrado com as taxas mensais e a dependência da nuvem de ferramentas de IA para reuniões, criou o Oats AI. É um assistente de reunião de IA gratuito, local e de código aberto, que funciona na máquina do usuário sem assinaturas.