18 steps. From a chat loop to a full AI agent. Every line commented. Build Your Own OpenClaw walks you through building an AI agent from scratch. Tool use, memory, skills, cron, multi-agent routing, web dashboard. Each step is runnable code with a README that explains the design decision behind it. Phase 1: a working single agent. Phase 2: event-driven refactor so it runs across platforms. Phase 3: autonomous scheduling and multi-agent coordination. 1.7K stars on GitHub. github.com/czl9707/build-you… Follow for daily dev finds 🔔

59K stars and barely anyone mentions it. Ruflo runs coordinated agent swarms on top of Claude, with memory and RAG baked in. Agents share what they learn, hand off work, and talk to Claude Code and Codex directly. Useful when you want agents that still know what they did the day before. ⭐ 59.3K #AI #OpenSource github.com/ruvnet/ruflo Follow for daily dev finds 🔔

Making Lottie animations means either paying for After Effects or hand-editing JSON until your eyes bleed. Diffusion Studio's text-to-lottie changes that. Install it as a Claude Code or Codex skill, describe the motion, timing, and camera moves in plain English, and you get a Lottie file with a built-in player for tweaking. Give it an SVG path and it animates the reveal. ⭐ 2.6K #Animation #DevTools github.com/diffusionstudio/l… Follow for daily dev finds 🔔

Locking your own engineers out of the AI they're building isn't security. It's self-sabotage. Anthropic got ordered to block all foreign staff. GPT-5.5 does the same thing. #AI axios.com/2026/06/12/anthrop…

When was the last time you read your oldest open issues? ClawSweeper scans every issue and PR on a schedule and tells you which ones are safe to close, and why. It writes one markdown report per item, only closes high-confidence proposals that haven't changed, and can open guarded fix PRs for the issues worth keeping. Think of it as a conservative janitor for your backlog. You host it yourself, so you fork the repo and point it at your own projects. ⭐ 1.8K #OpenSource #DevTools github.com/openclaw/clawswee… Follow for daily dev finds 🔔

4,756 files. Someone reverse-engineered Claude Code's whole source by pulling the sourcemap out of its npm package. The repo rebuilds 1,884 TypeScript files from cli.js.map, covering the tools, commands, services, the multi-agent coordinator, and even a vim mode. It's pinned to version 2.1.88, unofficial, and tagged research-only. If you've ever wanted to see how Anthropic's coding agent is wired under the hood, this is as close as you'll get without working there. ⭐ 9.4K #TypeScript #AI github.com/ChinaSiro/claude-… Follow for daily dev finds 🔔

Decks that don't look like default PowerPoint. Beautiful HTML Templates is a set of 34 slide decks built so any coding agent can clone one and hand you back a finished presentation. The agent reads an AGENTS.md file to match your brief to the right template. Think Cormorant Garamond on warm paper, forest-green editorial layouts, the works. It's all plain HTML too, so you can rewrite any of it. ⭐ 2.8K #Frontend #AI github.com/zarazhangrui/beau… Follow for daily dev finds 🔔

Palantir's contracts start at seven figures. This is the open version. Osiris pulls flight tracking, 2,000-plus CCTV feeds, earthquake data, and conflict-zone mapping into a single GPU-rendered map. It's built on Next.js 16 and MapLibre GL, so everything runs in WebGL and holds 60fps with thousands of things on screen at once. MIT-licensed, and there's a live demo you can poke around in. ⭐ 5.4K #OSINT #OpenSource github.com/simplifaisoul/osi… Follow for daily dev finds 🔔

Browser automation for AI agents usually means shipping a whole Playwright runtime plus 200MB of Node. agent-browser skips all of that. It's a native Rust CLI that pulls its own Chrome for Testing build, so the daemon needs no Node and no Playwright. If you already have Chrome, Brave, or Puppeteer installed, it just finds them. Vercel shipped it and the 36K stars tell you people were tired of the old way. ⭐ 36.0K #Rust #DevTools github.com/vercel-labs/agent… Follow for daily dev finds 🔔

Ever fed your whole repo to an AI reviewer and watched it eat 100K tokens just to touch two lines? Yeah, same. code-review-graph builds a map of your codebase that any MCP tool can query. The AI asks the graph for the files, functions, and deps it actually needs instead of cramming everything into context. Real reviews. Smaller context windows. Works with Claude Code, Cursor, anything MCP. ⭐ 18.5K #AI #DevTools github.com/tirth8205/code-re… Follow for daily dev finds 🔔

Your Mac already has an LLM in it. You're dropping $20/mo to talk to a server farm. apfel turns Apple Intelligence's FoundationModels into a CLI tool, an OpenAI-compatible server, and a REPL. All on-device. No API keys, no cloud, nothing to download. Pipe text through it. Hit localhost:11434/v1 with any OpenAI SDK. Or just chat. Tool calling works too. Needs macOS 26 Tahoe and an M1 or better. ⭐ 5.8K #Apple #AI github.com/Arthur-Ficial/apf… Follow for daily dev finds 🔔

Most multi-agent frameworks are just chatbots passing notes. Edict is different. It stole the accountability model from a 1300 year old government system (Tang Dynasty, three departments) and ran with it. 12 agents push work through planning, review, and execution. Any one of them can reject another's output before it ships. There's a live dashboard so you can watch the whole chain break or hold together in real time. More structured than CrewAI. More transparent than AutoGen. ⭐ 16.1K #AI #MultiAgent github.com/cft0808/edict Follow for daily dev finds 🔔

Watching Claude Code eat your context window with zero idea what's happening is the worst. claude-hud fixes exactly that, and 25K stars says a lot of us hit the same wall. It's a plugin that lives right below your input and shows context usage, active tools, running agents, and todo progress live. Two commands to install. Now you can actually see what's burning your tokens before you hit the ceiling. ⭐ 25.1K #ClaudeCode #DevTools github.com/jarrodwatts/claud… Follow for daily dev finds 🔔

WeChat killed its web client years ago. Now juggling sessions across your laptop, desktop, and phone is a daily headache. Some dev fixed it by running the full desktop client inside Docker and streaming it to any browser. WechatOnCloud hosts multiple WeChat instances on your NAS or server. Each one is a container running Xvfb the official client, piped to the browser via KasmVNC. Multiple people can share a single session across devices. Doesn't touch the client itself. ⭐ 2.5K #OpenSource #SelfHosted github.com/Gloridust/WechatO… Follow for daily dev finds 🔔

Anthropic's "too dangerous to release" model made a sheep herding game for its first real test. 45 minutes, €20 in tokens, 2,319 lines in one shot. Every prior model failed this benchmark. #AI koenvangilst.nl/lab/claude-f…

Writing a 2000-word article with zero images is a great way to lose readers. ian-xiaohei-illustrations reads Chinese articles and turns key concepts into hand-drawn illustrations. Not generic stock art. It finds the actual metaphor in your text, then draws it. 16:9 black ink on white, small red-orange-blue annotations. Generates a shot list and final PNGs. It's a Codex Skill built for knowledge-heavy posts where one diagram beats three paragraphs of explanation. ⭐ 4.1K #AI #CreativeCoding github.com/helloianneo/ian-x… Follow for daily dev finds 🔔

Keeping 15 browser tabs open to track news, markets, and infrastructure status is not a strategy. worldmonitor pulls all of it into one real-time dashboard. AI-powered news aggregation, geopolitical monitoring, infrastructure tracking. Variants for tech, finance, and commodities. Self-hosted, TypeScript, AGPL. 56K stars. The UI is clean enough to leave on a second monitor all day. ⭐ 56.4K #OpenSource #Dashboard github.com/koala73/worldmoni… Follow for daily dev finds 🔔