Dev Digest — April 11, 2026

🔥 HOT RELEASES

GLM-5.1 by Z.ai — First Open Model at Frontier Level The open-source model just hit #3 in Code Arena, surpassing Gemini 3.1 and GPT-5.4, now on par with Claude Sonnet 4.6. Running at ~6.5 t/s on M5 Max 128GB. One user reverse-engineered an entire compiled TypeScript app in a single 20M token shot. The open-source AI race just got real. 🔗 X: https://x.com/arena/status/2042611135434891592 🔗 X: https://x.com/arena/status/2042643933768151485 💻 Dense model export: https://github.com/Anemll/anemll-flash-llama.cpp/tree/GLM-Dense-Direct-slot-writes

Cursor Composer 2 — Double Usage Weekend Cursor is doubling Composer 2 usage this weekend in their new interface. No hourly limits. 🔗 X: https://x.com/cursor_ai/status/2043009335161430187

Vercel v0 — Work on Any GitHub Repo v0 now lets you open and work on any GitHub repo directly, from anywhere. No local clone needed. 🔗 X: https://x.com/vercel_dev/status/2042715005733474413

Shopify AI Toolkit Shopify launched an AI Toolkit letting merchants manage stores using Claude Code, Codex, Cursor and VS Code agents. Backend writes, product/order/inventory/SEO management all via prompts. 🔗 X: https://x.com/Cointelegraph/status/2042741940957905304

Claude Code /ultraplan New command that builds implementation plans in the web UI, lets you edit them, then runs execution back in terminal. 🔗 X: https://x.com/RoundtableSpace/status/2042728076765704251

Claude Code /advisor Anthropic's new advisor tool: Opus as supervisor, Sonnet/Haiku as executors. SWE-bench +2.7pt, costs down 11.9%. Haiku+Opus scored 2x on BrowseComp at 85% less than Sonnet. 🔗 X: https://x.com/masahirochaen/status/2042753248352964949

🧪 INTERESTING REPOS

OpenMontage — Agentic Video Production 11 pipelines, 49 tools. Describe a video in plain language → AI researches, scripts, generates assets, edits, renders. Full product ad for $0.69. Supports 12 video gen providers, 8 image gen, 4 TTS. Works with zero API keys using local models. 🔗 X: https://x.com/heygurisingh/status/2042811095703031969 💻 GitHub: https://github.com/calesthio/OpenMontage

Caveman — Save 65% Tokens in Claude Code A skill that makes your AI agent respond in "unga bunga" caveman mode while keeping technical precision. Up to 75% fewer output tokens. 🔗 X: https://x.com/midudev/status/2042964909441368497 💻 GitHub: https://github.com/JuliusBrussee/caveman

RTK — Token Compression for AI Coding Compresses command output (git, cargo, docker, kubectl, pytest) before feeding it to AI. Saves 60-90% of tokens from tool output noise. Works with Claude Code, Cursor, Gemini CLI, Copilot. 🔗 X: https://x.com/GitHub_Daily/status/2042885064917422318 💻 GitHub: https://github.com/rtk-ai/rtk

claude-mem — Persistent Memory for Claude Code Saves context between sessions automatically, recovers only what's needed. 47k+ stars. No more "this project is..." every session. 🔗 X: https://x.com/_guillecasaus/status/2042959633384423506 💻 GitHub: https://github.com/thedotmack/claude-mem

hermes-hudui — AI Agent Consciousness Dashboard Web UI for monitoring Hermes AI agents: live memory state, token costs with trends, skills, cron jobs, growth deltas. 4 cyberpunk themes including Blade Runner and fsociety. 🔗 X: https://x.com/sharbel/status/2042519782688211079 💻 GitHub: https://github.com/joeynyc/hermes-hudui

Multica — AI Kanban for Coding Agents Turn AI coding assistants into virtual team members. Assign tasks on a kanban board, AI auto-picks them up, reports blockers, updates status. Supports Claude Code, Codex. Docker self-host. 🔗 X: https://x.com/GitHub_Daily/status/2042837605725896878 💻 GitHub: https://github.com/multica-ai/multica

OpenScreen — Open-Source Screen Recorder for Product Demos 🔗 X: https://x.com/tom_doerr/status/2042669325409022271 💻 GitHub: https://github.com/siddharthvaddem/openscreen

claude-token-efficient — CLAUDE.md for Reducing Verbosity Drop-in CLAUDE.md that reduces output verbosity by ~63%. Kills the "Great question!" and over-engineered code patterns. 🔗 X: https://x.com/CamilleRoux/status/2042860205487083527 💻 GitHub: https://github.com/drona23/claude-token-efficient

RetriEval — Vector Search Benchmarks in Rust Benchmarking USearch, FAISS, cuVS, Weaviate, Redis, Qdrant from Rust. Covers scalar/binary quantization. 🔗 X: https://x.com/ashvardanian/status/2042746489542254693 💻 GitHub: https://github.com/ashvardanian/RetriEval

awesome-llm-apps — 100+ AI Agent Templates 105k+ GitHub stars. Agents, multi-agent teams, RAG templates. All free and open source. 🔗 X: https://x.com/Saboo_Shubham_/status/2042650161005121741 💻 GitHub: https://github.com/Shubhamsaboo/awesome-llm-apps

🎥 WORTH WATCHING

GLM-5.1 First Impressions — LM Arena Peter from LM Arena walks through the first frontier-level open model. 🔗 https://www.youtube.com/watch?v=f11tVBXWr2g

Knowledge Management System in Claude Code — Allie Miller Building a second brain with Obsidian + Claude Code + GitHub. Full demo. 🔗 X: https://x.com/alliekmiller/status/2042728780847047131 🎬 https://www.youtube.com/watch?v=v1suC9lWJ2I

Claude Code Pro Usage Full Course Comprehensive course on building apps and teams with Claude Code. 🔗 X: https://x.com/eng_khairallah1/status/2042680409306402998

Claude Code on Rooted Android — Reverse Engineering a Game Agent autonomously takes over a rooted phone and reverse engineers a mobile game. 🔗 X: https://x.com/RoundtableSpace/status/2042826223466754420

💡 TECHNIQUES & IDEAS

Token Saving Meta: Three separate projects dropped today all solving the same problem — AI agents burning too many tokens. Caveman (speak less), RTK (compress tool output), and claude-token-efficient (reduce verbosity). The token efficiency movement is here.

Obsidian as AI Second Brain: Allie Miller's setup — Obsidian for knowledge graph, Claude Code for querying/updating, GitHub for sync. The "second brain" concept finally has practical tooling.

Claude Code /insights: Run /insights to get Claude's analysis of your workflow — suggested CLAUDE.md updates, hooks, skills, and prompting advice. Hidden gem most people don't know about. 🔗 X: https://x.com/engineering_bae/status/2042687351978356916

Google Colab Learn Mode: Google added Custom Instructions + Learn Mode to Colab. Instead of just giving you code, it teaches you why. Flips the copy-paste pattern. 🔗 X: https://x.com/aiwithmayank/status/2042916369784803695

🔮 EMERGING TRENDS

Open Models at the Frontier: GLM-5.1 is the inflection point. First open model competitive with Claude Sonnet 4.6 in coding. Running locally on consumer hardware. The $0/mo coding stack (GLM-5.1 + OpenClaw/Claude Code) is now viable.

Agent Orchestration Going Mainstream: Multica (kanban for agents), OpenMontage (video production agents), hermes-hudui (agent monitoring) — we're moving from "one agent, one task" to "manage a fleet of specialized agents." The tooling is catching up to the vision.

Token Economy as First-Class Concern: Multiple viral projects this week focused purely on reducing token costs. As agents run longer autonomous sessions, token efficiency becomes as important as the model quality itself.

Claude Code Ecosystem Explosion: /ultraplan, /advisor, /insights, persistent memory, 27-agent configs, Shopify integration — Claude Code is becoming a platform, not just a tool. The ecosystem around it is growing faster than the tool itself.

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