Coding agents that combine models, runtimes like Claude Code, and tools such as file write carry inherent safety risks, including potential data loss or system damage from misaligned model behavior or…
AITerm, a native macOS terminal with an AI command loop and safety gate, is now available. The app lets users describe tasks in plain English, proposes editable shell commands for approval, and can fi…
A developer built GEO Auditor, an Apify Actor that checks a website's visibility to AI chatbots like ChatGPT, Claude, and Gemini. After discovering their own site scored 45/100, they found common issu…
Kimi K2, a MoE model with 1 trillion total parameters and 128k context, is available via Novita API at $0.57 per million input tokens and $2.30 per million output tokens, with 99.32% uptime. The model…
Atuin has open-sourced the Atuin AI server, allowing users to self-host the terminal-focused AI agent that provides agentic tools directly in the shell. The server supports any OpenAI-compatible endpo…
Agent loops are often oversimplified as a single loop, but they actually consist of three distinct loops: the inference loop, the tool loop, and a third loop not detailed in the excerpt. The inference…
Chinese open-weight models accounted for 41% of downloads on Hugging Face this spring, surpassing U.S. models, and the top six most popular models on OpenRouter are all open models from Chinese firms …
CrowdMind, a free desktop app for building synthetic research panels with AI personas, has been released by developer Brokenwatch24. The local-first tool lets founders, product marketers, and research…
Reflect, an open-source note-taking app built around a folder of Markdown files, has been released for Mac, iOS, and Chrome. The app offers daily notes, wiki links, local search, and optional AI query…
Hayden Bleasel, an expert developer from OpenAI, released Blume, an open-source, zero-config documentation framework that ships AI-ready docs from a Markdown folder. Blume version 1.0.3 is available o…
A developer tested over 300 AI models on real-world agent coding tasks, then killed the benchmark because it was no longer useful. The leaderboard showed most models performing well at low cost, but t…
A new study from arXiv introduces the Format Sensitivity Index (FSI) and Parseability Sensitivity Index (PSI) to measure how prompt wrapper formatting affects LLM benchmark scores. Across 140,000 Open…
Usage of U.S. AI models fell in the first week of July, and Chinese AI usage surpassed U.S. AI by about 40%, according to OpenRouter data cited by Deutsche Bank chief economist Jim Reid. Dean Baker, s…
A new free and open-source browser extension called Waffy lets users automate web tasks and chat about page content using AI models from OpenAI, Anthropic, Google, xAI, Groq, or OpenRouter, with keys …
Kairos, an experimental local-first AI agent platform, has released an early MVP featuring goal mode, guarded swarms, smart model routing, memory, tools, and a bundled AI skills library. The terminal-…
Canvas, a multiplayer HTML editor for decks that is agent-agnostic and MCP-first, has been released as a standalone tool with its own Supabase project and auth. The editor supports slide ownership by …
Consortium, an open-source workflow runtime for LLM-powered applications, enables developers to replace a single model call with reliable, auditable ensemble workflows by changing one line of code. Th…
AI models retain between high single digits and 40% of customers after five months, with the stickiest foundational cohorts near the top of that range, according to a study by OpenRouter and a16z. The…
Developer Martin Piaggi released Summarize, an open-source tool that converts videos from YouTube, TikTok, Instagram, and other sources into structured notes using AI transcription and LLM summarizati…
Jules Storer, creator of the JUCE audio framework, released Juggler, an open-source GUI coding agent that provides a visual workbench for inspecting and controlling LLM interactions with a codebase. T…