All published AI Radar articles.

HeyClaude is a GitHub-based curated registry and distribution surface for Claude and AI-workflow assets, including agents, MCP servers, skills, commands, hooks, rules, guides, tools, jobs, Raycast feeds, and static data exports. It also provides an npm MCP package.

HarnessClaw Engine is an open-source LLM programming assistant engine built with Go. It supports WebSocket for multi-turn dialogues, tool calling, permission control, and skill extension, and recently released version v0.0.18.

Endava is integrating AI agents, ChatGPT Enterprise, and Codex to enhance software delivery, automate workflows, and cultivate an AI-centric organizational culture.

The `awesome-ai-agents-2026` GitHub repository provides a curated list of AI agent frameworks, tools, platforms, and resources. It aims to be a definitive guide for the year 2026, which the repository's description labels as the year AI agents went mainstream.

A new research paper investigates the use of transformer model activations for selecting in-context examples in large language models. The study, which includes a comprehensive analysis using Llama-3.2-3B and Qwen2.5-3B, found that MLP outputs based on massive activations or statistical moments do not correlate with example quality or task performance, suggesting that this approach is not effective for in-context learning.

GeM-NR is a new training-free method for multi-view consistent image editing, specifically designed to handle nonrigid scene changes that significantly alter geometry and appearance. It addresses limitations of existing methods that primarily focus on rigid or appearance-only edits.

Open-Source-Legal has developed 'cite', a Python-based GitHub project designed to create a "ground truth layer" and version control system for knowledge, facilitating collaboration between humans and AI agents. It aims to build an open citation graph for document repositories, allowing agents to traverse relationships between documents and propose new annotations.

Researchers have introduced DistIL, a new approach to reinforcement learning that leverages rich feedback beyond simple binary rewards. DistIL uses a distributional variant of the DAgger imitation learning algorithm with a forward cross-entropy objective, which allows for more effective credit assignment and guarantees monotonic policy improvement. This method has shown empirical improvements over traditional RL from verifiable rewards (RLVR) and self-distillation baselines in tasks like scientific reasoning, coding, and complex mathematical problem-solving.

A new text generation model, `rupeshpanda/gita-text-generation-gpt2`, has been identified on Hugging Face. Built on the GPT-2 architecture and utilizing the Transformers library, this model is designed for text generation tasks.

Wasmer utilized OpenAI's Codex and GPT-5.5 to develop a Node.js runtime optimized for edge environments. This collaboration reportedly accelerated development significantly, reducing project timelines from months to weeks.