Tag
The paper introduces Phronix, a capability-centric learning paradigm that lets LLM-based agent harnesses acquire, validate, and reuse capabilities from their own executions via a predictor-validator-corrector loop, achieving 93-95% first-attempt success on recurring tasks and boosting cheaper models from ~28% to ~90% on SWE-bench.
Ogcode 的团队分享了其用户数据洞察：开发者安装大量 skills 和 MCP 工具导致 agent 选错工具并浪费高达 77% 的 token，他们正在研发能按需动态选择工具的解决方案，并向社区征集思路。
Ars Contexta outlines how to build agent repos and knowledge systems that give AI agents persistent context, domain-specific tools, and reusable working methods for knowledge work, arguing for custom apps built on your own data instead of disconnected systems of record.
Introduces KoboldCpp Agent, a built-in agentic harness for easy code creation and general tasks with support for MCP tools and third-party backends, while also alerting to a fake phishing site targeting users.
BrowserClaw is a lightweight Chrome automation platform that uses a dual-brain architecture to optimize browser interactions for AI agents, significantly reducing latency and token usage.
The tweet advises using MCP tools for building custom agent harnesses because frontier LLMs are highly familiar with MCP, making testing and integration easier, based on the author's positive experience.
Conduct is an open-source runtime governance system for AI agents that enforces policies before actions execute, with signed configuration and hash-chained audits for LLM and MCP tool calls.
The author built an open-source harness for running local LLMs using Qwen 27b, featuring just-in-time code review, sub-agents, voice dictation, and more, and shares development insights.
A developer reconstructed Grok Bot 0.18 into readable TypeScript and created a shell that routes tasks through AI tools like Claude Code and Codex, with local execution and plugin support.
This paper proposes VibeWorlding, a framework for benchmarking and training multimodal agents to construct 3D open worlds from user queries, showing that reinforcement learning improves open-source models to compete with closed-source frontiers.
Aimock is an open-source library for mocking AI components like LLMs, tools, agents, and databases, enabling end-to-end testing without real implementations. It has achieved 1 million weekly installs, indicating its utility in the AI development community.
Open-sourced a desktop workspace that provides a governed runtime for AI coding agents, offering 100+ MCP tools, RBAC, and a self-evolving toolbox.
Fay is a fully open-source, commercially free digital human framework that supports offline operation, modular design, MCP tool invocation, and bionic memory. It features a built-in web management interface, enabling quick setup for virtual streamers, intelligent customer service, personal AI assistants, and more.
A user demonstrates giving a local LLM agent MCP tools for local image and video generation, enabling fully offline and free generation on demand.
Lemonade v10.8 introduces auto memory management, cloud offload, improvements to Omni, and the ability to call local AI models as MCP tools.
Pepti-Agent is a closed-loop AI framework for therapeutic peptide design that uses MCP tools and an LLM controller to iteratively refine sequences based on multi-property profiles, addressing constraints like solubility, hemolysis, and non-fouling.
DeerFlow 2.0 adds multi-workspace isolation, multi-role permission management, and workspace-level MCP tool and skill isolation on top of the original open-source project. It supports independent chat history, memories, and tool sets. The backend uses FastAPI + LangGraph, and the frontend uses Next.js 16.
Repowise is an open-source tool that indexes codebases into five intelligence layers—dependency graph, git history, auto-generated docs, architectural decisions, and code health—and exposes them to AI coding agents via MCP tools for more accurate context and fewer tool calls.
Reachy Mini's conversation app can now use tools hosted in Hugging Face Spaces via MCP, allowing users to add abilities like weather checking or web search with a single command.
DeepSeek-Code-Whale is an open-source terminal AI coding agent, specifically optimized for DeepSeek models, supporting MCP tools, Skills extensions, prefix caching optimization (90% cache hit rate) and 1M context window, aimed at reducing AI coding costs and providing efficient command-line workflows.