Industry

vLLM v0.25.0 makes Model Runner V2 the default for all dense models

vLLM shipped v0.25.0 with 558 commits from 232 contributors, and Model Runner V2 is now the default execution path for all dense models. This is the inference engine behind a huge share of self-hosted and enterprise LLM deployments, so a default runner swap is a real performance and stability shift for anyone serving open models. If you run models yourself, this release changes your baseline.

Model

Claude Code v2.1.207 turns on Auto mode by default for Bedrock, Vertex AI, and Foundry

Anthropic's latest Claude Code release removes the opt-in flag for Auto mode on Bedrock, Vertex AI, and Foundry, meaning enterprise cloud deployments now get autonomous mode out of the box unless admins disable it in settings. That is a quiet but meaningful shift in how much autonomy coding agents get by default inside big-company environments. Worth knowing if you advise teams on agent rollout or governance.

Industry

Who does Anubis actually stop? A hard look at the popular AI-crawler blocker

Anubis is the proof-of-work challenge tool many sites deployed to block AI scrapers, and this analysis digs into who it actually keeps out versus who walks right past it. The short version: determined AI crawlers with headless browsers pay the cost easily, so the protection is thinner than the adoption suggests. Directly relevant if you think about bot defense, scraping economics, or the arms race between publishers and AI labs.

Industry

Graphify: turn any folder of code, schemas, and docs into a queryable knowledge graph

Graphify is an AI coding assistant skill that ingests a folder of code, SQL schemas, shell scripts, docs, papers, even images and video, and builds one queryable knowledge graph across app code, database schema, and infrastructure. It plugs into Claude Code, Codex, OpenCode, Cursor, and Gemini CLI, so your agent can answer questions about how the whole system fits together instead of grepping file by file. It is the breakout repo of the moment at over 82,000 stars.

Industry

Addy Osmani's agent-skills: production-grade engineering skills for AI coding agents

Addy Osmani published a collection of production-grade engineering skills you can drop into AI coding agents, and it has already crossed 77,000 stars. Instead of writing your own prompts for code review, refactoring, or debugging discipline, you install skills that encode how a senior engineer actually works. A fast way to upgrade an agent setup you already run today.

Industry

open-design: an open-source, local-first design engine powered by your coding agent

open-design is a local-first desktop app that turns whatever coding agent you already use, Claude Code, Codex, Cursor, Gemini, and 20 plus CLIs via bring-your-own-key, into a design engine. It produces prototypes, landing pages, dashboards, slides, images, and video as real files with HTML, PDF, PPTX, and MP4 export. If you have been paying for a hosted design tool, this is the open alternative to try this week.

Industry

rag-rat: an MCP server that gives coding agents the history behind your code

rag-rat is a repository-intelligence MCP server for Claude Code and Codex that solves the cold-start problem: every new session, your agent has to re-learn which parts of the code are load-bearing, who calls what, and which alternatives were already rejected. It serves that accumulated context so the agent starts each session already knowing the why behind the code. Small install, and it targets one of the most annoying gaps in day-to-day agent work.

Industry

codex-ollama-proxy: keep Codex plugins and MCP tools working with local models

Codex can technically connect to Ollama and other Responses API providers, but native plugin tools, MCP namespaces, tool_search, and apply_patch often break with custom or local backends. codex-ollama-proxy is a local compatibility layer that flattens those gaps so the full Codex tool stack works against models you run yourself. If you want agent workflows without sending code to a hosted API, this is the missing shim.