โ† All digests

Dev Digest โ€” April 12, 2026 (Sunday)

๐Ÿ”ฅ HOT RELEASES

Bun: WebSocket Unix Domain Sockets Next version of Bun adds native WebSocket client support for Unix domain sockets. Huge for local service communication. ๐Ÿ”— X: https://x.com/bunjavascript/status/2043179050664169973

Hermes Agent hits 62K stars NousResearch's Hermes Agent gained 6,438 stars in 24 hours alone โ€” now at 61.8K total. "The agent that grows with you" is the fastest-growing agent framework right now. ๐Ÿ”— X: https://x.com/trending_repos/status/2043252219374670103 ๐Ÿ’ป GitHub: https://github.com/NousResearch/hermes-agent

Distill Desktop โ€” General Availability Distill Desktop is now available to everyone. Local-first AI distillation tool. ๐Ÿ”— X: https://x.com/TGUPJ/status/2043378375377432604

GitButler "Plain Git" Mode (almost ready) GitButler is close to shipping a "plain git" mode โ€” mix GitButler with regular git commands in any repo without the workspace checkout requirement. ๐Ÿ”— X: https://x.com/gitbutler/status/2043252948394770846 ๐Ÿ’ป GitHub: https://github.com/gitbutlerapp/gitbutler/issues/11866


๐Ÿงช INTERESTING REPOS

claude-code-best-practice โ€” #1 Trending on GitHub 84 sourced tips, implementation examples for every major Claude Code feature, workflow comparisons across 8 repos, and tips from the creator of Claude Code. 19.7K stars. ๐Ÿ”— X: https://x.com/sukh_saroy/status/2043199144429531433 ๐Ÿ’ป GitHub: https://github.com/shanraisshan/claude-code-best-practice

andrej-karpathy-skills Production-grade engineering skills for AI coding agents based on Google's best practices. 20 skills, 7 slash commands. Works with Claude Code, Cursor, Gemini CLI. ๐Ÿ”— X: https://x.com/Saboo_Shubham_/status/2043172313475211273 ๐Ÿ’ป GitHub: https://github.com/addyosmani/agent-skills

Memory Sparse Attention (MSA) Supports 100M token context on just two A800 GPUs. Dual routing attention + KV cache compression. Hit 3K stars fast. ๐Ÿ”— X: https://x.com/elliotchen100/status/2043255919409254865 ๐Ÿ’ป GitHub: https://github.com/EverMind-AI/MSA

PLFM_RADAR โ€” Open Source Military Radar Full open-source phased array radar tracking targets up to 20km. PCBs, FPGA code, schematics, Python GUI. MIT license. Contractors charge $250K for this. ๐Ÿ”— X: https://x.com/HowToAI_/status/2043340557926477832 ๐Ÿ’ป GitHub: https://github.com/NawfalMotii79/PLFM_RADAR

crawl4ai โ€” LLM-friendly Web Crawler Purpose-built crawler that outputs clean, structured content optimized for LLM consumption. ๐Ÿ”— X: https://x.com/tom_doerr/status/2043215600852357574 ๐Ÿ’ป GitHub: https://github.com/unclecode/crawl4ai

b4n โ€” k9s-inspired Kubernetes TUI in Rust Terminal UI for Kubernetes built with kube-rs and ratatui. Full CRUD, log streaming, shell access, port forwarding. ๐Ÿ”— X: https://x.com/K8sArchitect/status/2043042912322109849

DeepTutor by HKUDS Top 5 GitHub repos this week โ€” AI tutoring system. ๐Ÿ’ป GitHub: https://github.com/HKUDS/DeepTutor

Multica Multi-agent AI platform, trending on GitHub this week with hundreds of updates. ๐Ÿ”— X: https://x.com/jiayuan_jy/status/2043201759942971864 ๐Ÿ’ป GitHub: https://github.com/multica-ai/multica

Lunel โ€” Open Source iOS App 100% open source, completely free iOS app. ๐Ÿ”— X: https://x.com/damnGruz/status/2043061983252222331 ๐Ÿ’ป GitHub: https://github.com/lunel-dev/lunel

Zen Browser โ€” 41K Stars Open-source Chrome alternative built on Firefox. Glance (modal link previews), workspaces, split view, compact mode, 79+ mods. ๐Ÿ”— X: https://x.com/toolfolio/status/2043200224777441486

gbrain by Garry Tan Open source personal AI software โ€” MIT licensed, community of homebrew builders. ๐Ÿ”— X: https://x.com/garrytan/status/2043024918363975911 ๐Ÿ’ป GitHub: https://github.com/garrytan/gbrain


๐Ÿ’ก TECHNIQUES & IDEAS

Mind โ€” Persistent Memory Across All AI Coding Agents MCP server backed by SQLite that gives persistent memory to Claude Code, Cursor, Codex, Gemini CLI, and Windsurf. 4-tier memory model (hot/warm/cold/frozen), full-text search via FTS5, tags + graph links between memories. ๐Ÿ”— X: https://x.com/heygurisingh/status/2043219220146897358

OpenAI vs Anthropic: Two Paths to Agent Harness Excellent analysis: OpenAI bets on full automation (Codex replacing devs), Anthropic bets on collaboration (Claude Code as co-pilot) then absorbing the harness layer into "Managed Agents." Both racing to become the Agent OS. ๐Ÿ”— X: https://x.com/Yonah_x/status/2043230047696572843

Claude Code Regression Data AMD's Senior AI Director analyzed 6,852 real coding sessions (Jan-Mar 2026). Found measurable decline: thinking depth dropped, model reads less of codebase before changes, skips files on ~1/3 of edits, defaults to simpler fixes. ๐Ÿ”— X: https://x.com/TFTC21/status/2043031112591773745

"Vibe Coding Only Works If You Already Know How to Code" Hot take that resonated โ€” the distinction between vibe coding (useful prototyping) and shipping code you don't understand to production. ๐Ÿ”— X: https://x.com/KaiXCreator/status/2043221320226488662

tsgo Excitement Once tsgo (TypeScript's Go-based type checker) stabilizes, agent typechecking loops will be dramatically faster. ๐Ÿ”— X: https://x.com/kennethnym/status/2043289733145469085


๐Ÿ”ฎ EMERGING TRENDS

Agent Memory is the New Frontier โ€” Mind, MSA, and claude-code-best-practice all tackle the same problem: AI agents forget everything between sessions. SQLite-backed memory layers, 100M token context windows, and curated CLAUDE.md files are the competing approaches.

The Harness Wars โ€” OpenAI (Symphony/Codex) vs Anthropic (Managed Agents) are racing to own the agent runtime layer. Whoever controls the harness controls the developer workflow.

Claude Code Ecosystem Exploding โ€” 5 of the top trending GitHub repos this week are Claude Code-related. The ecosystem of skills, workflows, and best practices is maturing fast.

Open Source Hardware โ€” Military-grade radar systems showing up on GitHub under MIT license. The open-source ethos is pushing into physical hardware in ways that would've been unthinkable 2 years ago.

Top AI Papers of the Week (via @dair_ai): Memento, Neural Computers, The Universal Verifier, Agent Skills in the Wild, Memory Intelligence Agent (MIA), Single-Agent vs Multi-Agent LLMs, Scaling Coding Agents via Atomic Skills. ๐Ÿ”— X: https://x.com/dair_ai/status/2043354582319870362


Full digest: https://digest.99.cooking