News this week
New AI and agent tool releases, updated daily from X, Product Hunt and GitHub Trending. Built from data/news.json.
AI Gateway logs now group requests into sessions and tag each with the coding agent or app that ran it (Claude Code, Codex, Cursor), with cost, models, providers and cache hit rate per session. Session affinity keeps each turn on the same provider so more of the prompt is read from cache.
Link · Source: Vercel changelog, Oct 18 (via X)
Open-source (Apache 2.0) hook and local proxy for Claude Code, Codex, Copilot CLI and VS Code. In Claude Code it masks secrets and personal data in tool results before the model reads them; in the others it mostly reports or blocks. macOS and Linux only. #3 on Product Hunt for the day.
Link · Source: Product Hunt daily leaderboard, Oct 18 + product page
AI on-call engineer that runs inside your network: it checks live systems to confirm the root cause of an incident, dry-runs a fix, and applies it after one human approval in Slack, PagerDuty or similar, with a hash-chained audit log. #7 on Product Hunt for the day.
Link · Source: Product Hunt daily leaderboard, Oct 18 + product site
Open-source AI email assistant and email client, pitched as a Superhuman alternative. An existing project (~12.7k GitHub stars) that relaunched on Product Hunt and finished #1 for the day.
Link · Source: Product Hunt daily leaderboard, Oct 18 + GitHub repo
Elasticsearch-compatible search index for agents: one command indexes code, docs, logs and PDFs so agents can search without grep. GitHub Trending ~1,648 stars today.
Link · Source: GitHub Trending (daily)
Add-ons for coding agents like Claude Code, Codex, OpenCode and Cursor covering skills, memory, security and research-first workflows. GitHub Trending ~1,057 stars today.
Link · Source: GitHub Trending (daily)
Job search framework built on Claude Code that runs on your own machine: evaluates postings, tailors CVs, writes cover letters and preps interviews. GitHub Trending ~898 stars today.
Link · Source: GitHub Trending (daily)
Updated open-weight DeepSeek V4.1 Flash model (MIT license) with better coding and agent scores and shorter outputs. DeepSeek's own numbers put Terminal-Bench 4.0 at 42.3, up from 31.2. Live in DeepSeek's app and API and on Ollama Cloud.
Link · Source: @deepseek_ai on X + Hugging Face model card
Platform for building MCP servers and AI apps: open-source TypeScript and Python SDKs, a visual testing studio, workflow orchestration, and hosted deploys that connect to ChatGPT, Claude and other clients. #1 on Product Hunt for the day (~292 votes).
Link · Source: Product Hunt daily leaderboard, Oct 17
Open-source (Apache 2.0) desktop app from Craft for working with agents locally. Bring your own model provider, connect tools through MCP and APIs, and run scheduled automations. Launched on Product Hunt Oct 17 (#7); the repo has ~7.3k GitHub stars.
Link · Source: Product Hunt daily leaderboard, Oct 17 + GitHub repo
Open-weights mixture-of-experts model (816B total / 42B active), reportedly optimized to run locally on a Mac Studio.
Link · Source: Context Pill citing TestingCatalog on X
Agentic skills framework and software-dev methodology for coding agents; GitHub Trending ~973 stars that day.
Link · Source: GitHub Trending via Context Pill
Anthropic-managed directory of Claude Code plugins; GitHub Trending ~1,552 stars that day.
Link · Source: GitHub Trending via Context Pill
Reflection's first open-weight model: a 501B-parameter mixture-of-experts (23B active per token) built for coding and agent work, with a 1M-token context, released under Apache 2.0 with weights on Hugging Face.
Link · Source: Reflection blog and Hugging Face blog (Oct 16); X News
Visual reasoning model that follows objects through motion and changing viewpoints in images and video. Available as a research preview through Elorian's API and chat app.
Link · Source: Elorian blog (Oct 16); X News
First release of an open stack for computer-use agents that mix screen control with code execution, including prompts, evaluation runs from eight models, and early versions of a real-device harness and cross-device SDK.
Link · Source: AI Agent News; Hugging Face paper page
GPT-6 Pro in ChatGPT can now split a task across several agents working in parallel, with progress you can follow; OpenAI also says responses are faster and plugin/tool use is better. Rolling out to Pro subscribers.
Link · Source: @OpenAI on X (help-center release notes blocked curl with 403)
Persistent, forkable drives for Sandbox agents are now GA; Vercel Connect can forward MCP server webhook events to agents (public beta).
Link · Source: AI Agent News
177 science skills for agents like Claude Code and Codex.
Link · Source: GitHub Trending via Context Pill
Perplexity open-sourced the sandbox behind its Computer agent. It runs agent code in isolated Firecracker microVMs that apps control over an HTTP API. Apache 2.0.
Link · Source: Perplexity blog (Oct 15) and posts on X
Open-source generative UI framework: agents stream a compact UI language that renders as cards, forms and charts. 1.0 adds custom actions and functions; the 1.0 features need the React renderer for now.
Link · Source: X News and OpenUI docs; 1.0 npm packages published Oct 12, announced Oct 15
Lets AI agents pay by card and sign in to websites using placeholder tokens, so real card numbers and passwords never reach the agent. Includes spending caps, merchant allowlists and a kill switch; Free, Plus and Enterprise tiers.
Link · Source: Self Labs announcement via Business Wire (Oct 15); X News
Open-source (Apache 2.0) memory engine and Python SDK for agents, with separate pipelines for chat, working and coding memory, plus an open 30B-A3B memory model. MemHub is a local-first desktop app and CLI that pulls history from Claude Code, Cursor, Codex and others into one place for shared recall.
Link · Source: GitHub repo + PyPI (2.0.1 published Oct 15); widely shared on X Oct 18-19
Open-source (AGPL-3.0) task-focused agent workspace with native MCP support. Version 2.0 adds one-click install, visual configuration, better handling of long-running tasks and new Word, Excel and PowerPoint skills.
Link · Source: GitHub README (2.0 dated Oct 15); shared on X Oct 19
Fully local, open-source alternative to ElevenLabs for voice cloning, dubbing and transcription.
Link · Source: GitHub Trending via Context Pill
Open-source auditing of whether your AI agent does what it should.
Link · Source: GitHub Trending via Context Pill
Notion's update for using it as shared memory across agents: connect any MCP-compatible agent so it can read your workspace and update docs and task trackers, letting you switch agents without re-explaining your work. Launched on Product Hunt Oct 17 (#3).
Link · Source: Notion release notes (dated Oct 13) + Product Hunt Oct 17
Open-source TypeScript framework for agentic apps with agent-friendly UIs.
Link · Source: Builder.io blog (exact date not confirmed)
Open-source robotics framework where a coding agent writes, tests and repairs robot control code and saves working fixes to a reusable skill library. Apache 2.0; its docs warn the generated code is not sandboxed.
Link · Source: X News (Oct 16). Exact release date unclear: GitHub has no tagged v1.0 release, so month only
Get the Daily 5
The top 5 AI and agentic tool picks each day, straight to your inbox. Free.