Perplexity Computer now runs fully locally on a Mac

#1 𝕏

Aravind Srinivas announced that Perplexity Computer can now run fully locally on a Mac, without cloud-based inference or Wi-Fi.

#2 𝕏

Mistral AI announced Voxtral 3, describing its Transcribe and Speech models as state-of-the-art and as powering Mistral’s new voice agents API.

Also covered by: @Mistral AI

#3 📝 OpenAI News

OpenAI acquires Orchestra Bio - OpenAI announced on October 20, 2026, that it had acquired Orchestra Bio, a platform that helps biopharma teams coordinate the complex drug development process.

Also covered by: @OpenAI

#4 🐙 anthropics/knowledge-work-plugins

anthropics/knowledge-work-plugins - anthropics/knowledge-work-plugins is an Apache-2.0 repository shared by anthropics with 11 open-source plugins for Claude Cowork and Claude Code. Its file-based Markdown and JSON components bundle reusable skills, connectors, slash commands, and sub-agents across knowledge-work functions; the repo has 30,957 GitHub stars and 267 daily trending gains. Worth trying if you need customizable role-specific workflows without code, infrastructure, or build steps.

#5 🐙 android/skills

android/skills - android/skills is an Apache-2.0 repository of reusable, AI-optimized instructions that ground agents in Android development workflows and best practices. On October 21, 2026, android released Android Skills v1.0.16 with 3 new skills for on-device LLM apps, ANR auditing, and Android XR projected hardware, plus 13 updates; the repo has 9,097 stars and 1,255 weekly trending gains. Worth trying if you need specialized Android guidance for coding agents. Release notes

#6 🐙 cloudflare/security-audit-skill

cloudflare/security-audit-skill - cloudflare/security-audit-skill is an MIT-licensed coding-agent skill shared by Cloudflare for six audit phases, with specialized guides, a JSON schema, Node.js validators, machine-readable findings, and reports. It has 34,947 GitHub stars and 1,712 daily trending gains; Cloudflare says a single test run found roughly half the vulnerabilities that repeated runs found in total. Worth trying if you need repeatable codebase audits and have a tool-using, parallel-agent setup with Node.js and an OS-enforced sandbox.

#7 𝕏

Harrison Chase shared that LangChain’s managed deep agents builds on the deep-agents harness, or agent loop, by adding per-user credentials and memory when many people call the same agent.

#8 𝕏

Philipp Schmid shared that Gemini Managed Agents is the easiest way to try Gemini 4 Argon in an agentic environment. A single API call provides Python, Node.js, Git, Bash, full network access and control, a Credentials API, persistent filesystem, and background support.

#9 ▶️

Claude Code for creators: AI editing is here

How I AI Podcast

Also covered by: @Claire Vo

#10 𝕏

Anthropic announced a Science Blog post on designing protein binders against scorpion neurotoxins, an early step by Claude toward finding new treatments. About 1.2 million people are stung by scorpions annually, while antivenoms are often costly or in short supply.

Also covered by: @Anthropic

#11 🐙 Graphify-Labs/graphify

Graphify-Labs/graphify - Graphify-Labs/graphify maps code, docs, SQL schemas, configs, and PDFs into a queryable knowledge graph, with a reusable `/graphify` skill for Claude Code, Cursor, Codex, and Gemini CLI; code parsing is local and deterministic, without a vector store. Graphify-Labs released Graphify v0.9.93 on October 20, 2026, incorporating 22 community pull requests focused on declaration resolution and reliability fixes. Worth trying if you need explainable project context for AI coding assistants. Release notes

#12 🐙 eagle-hq/EAGLE-4

eagle-hq/EAGLE-4 - **eagle-hq/EAGLE-4** released v0.1.0 on October 20, 2026, providing inference code and the `eagle-harness` plugin for adaptive speculative decoding with patched SGLang, vLLM, and experimental TokenSpeed builds. The repo includes documentation, tests, Docker assets, and reported SGLang/vLLM speedups of 1.35×–2.99×; worth trying if you need workload- and hardware-adaptive LLM inference acceleration. Release notes

#13 🐙 K-Dense-AI/scientific-agent-skills

K-Dense-AI/scientific-agent-skills - K-Dense-AI/scientific-agent-skills is an MIT-licensed library of 177 reusable research skills and access guidance for 100+ scientific databases, spanning biology, medicine, engineering, Earth science, data analysis, and more. Each skill covers a package, data source, or workflow, and the library works with compatible hosts including Cursor, Claude Code, Codex, and Google Antigravity. Worth trying if you need portable scientific workflows for an AI agent.

#14 🐙 stablyai/orca

stablyai/orca - stablyai/orca is an MIT-licensed app for running CLI coding agents across desktop, mobile, and remote environments; its repo includes reusable skills, examples, tests, and cloud/mobile code. stablyai released Orca v1.4.237 on 2026-10-21 at 07:25:23Z, citing 100,000 GitHub stars; discovery metadata showed 100,407 stars and a 1,565 daily gain. Worth trying if you need to steer multiple coding agents across local and remote workspaces. Release notes

#15 🐙 wshobson/agents

wshobson/agents - **wshobson/agents** is an MIT-licensed collection for using shared plugins, agents, skills, and commands across Claude Code, Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, and Pi. Its catalog has 94 plugins, while the local source contains 202 agents, 185 skills, and 105 commands; worth trying if you need reusable coding workflows across multiple AI tools.

#16 🐙 thedotmack/claude-mem

thedotmack/claude-mem - **thedotmack/claude-mem** is an Apache-2.0 persistent-memory system that captures agent activity, summarizes it, and recalls relevant context across sessions; its repository includes plugins and integrations for multiple coding agents. thedotmack released Version 15.0.1 on 2026-10-20 at 22:40:29Z, aligning Smart search, timeline context, and observation details to one project scope while preserving the default three-step workflow and optional `dumb_search` from Version 15.0.0; the repository has 102046 GitHub stars. Worth trying if you need cross-session memory across supported agent environments. Release notes

#17 🐙 mvschwarz/openrig

mvschwarz/openrig - **mvschwarz/openrig** is Apache-2.0 software for persistent Claude Code and Codex teams defined in YAML; the repo contains packages, reusable skills, Docker files, scripts, tests, and architecture and security docs. mvschwarz released v0.6.14 on October 21, 2026, with stability fixes and in-place repair for older agents with unrecorded panes; it has 8,027 GitHub stars. Consider it if you need persistent coding-agent teams and can run Node.js 22 or 24 with tmux on macOS, Linux, or WSL2. Release notes

#18 🐙 ifixai-ai/iFixAi

ifixai-ai/iFixAi - **ifixai-ai/iFixAi** is an Apache-2.0 AI-agent auditing toolkit with a CLI, plugins, reusable skills, documentation, and examples. On October 21, 2026, ifixai-ai released v4.2.1, adding the report-only V11 inspection with 40 four-turn chats testing whether companion agents help vulnerable users harm themselves. Worth trying if you need structured safety checks alongside broader operational audits. Release notes

#19 𝕏

Harrison Chase commented that judges on production traffic are most useful when teams calibrate them on human labels, flag low-scoring outputs, confirm real failures, add them to evals, fix them, and rerun. Production traffic can uncover failures missing from offline datasets.

#20 𝕏

Guillermo Rauch announced an open-source, runtime-agnostic solution that extends the durability infrastructure powering Eve and Workflows to harnesses including Pi and OpenCode, with support for local, Vercel, or DIY deployments. He added that Claude Code and Codex can be made durable and observable this way, though their lack of embedding support makes them less efficient and portable and requires an extra sandbox compared with Pi or Fx.

Also covered by: @Guillermo Rauch