Mike Hutu

2/2 Most Electronic Shelf Label (ESL) systems are black boxes, making it incredibly difficult to study their communication patterns or signal integrity without expensive, specialized equipment. github.com/i12bp8/TagTink…

🧵 1/2 Manually intercepting and analyzing proprietary e-ink retail tag protocols has always been a massive friction point for hardware researchers.

2/2 That’s a community vote that a native macOS tool can beat the heavyweight IDEs for speed, memory footprint, and macOS‑centric shortcuts. github.com/coteditor/CotE…

🧵 1/2 If you’re sick of Electron‑bloat every time you need to tweak a config file, meet CotEditor. The Swift‑written plain‑text editor just hit 7.0.2 and it’s already racking up 8,001 GitHub stars.

2/2 For founders, the bar for “unicorn‑level” funding has moved, and investors are now asking tougher questions about safety and governance before writing the next big check.

🧵 1/2 Anthropic just blew past the $1 trillion mark in pre‑IPO valuation, a $200 B jump in five days. That’s not just hype – it reshapes the capital landscape and puts pressure on rivals to prove they can scale responsibly.

3/3 anthropic.com/constitution

2/3 In a crisp 30‑minute run they walk you through the charter’s core principles, the philosophical debates that shaped it, and why they chose a spoken format. No more skimming dense PDFs when you’re in the middle of a prompt‑tuning sprint.

🧵 1/3 Anthropic just dropped an audiobook version of Claude’s Constitution, narrated by its own architects Amanda Askell and Joe Carlsmith.

2/2 OpenBMB’s MiniCPM‑V hits that sweet spot: a Python‑based multimodal model that runs locally on mobile hardware, delivering image and video understanding with latency in the low‑hundreds of ms and a fraction of the cloud cost. github.com/OpenBMB/MiniCP…

🧵 1/2 The hype around massive cloud MLLMs is fading—developers need something that works offline, cheap, and fast.

2/2 The new v0.0.100 bundles 10+ state‑of‑the‑art models—ArcFace, VGGFace, Facenet, and even age/gender estimators—behind a single, lightweight Python API. No more version wars, no more hunting for compatible CUDA builds. github.com/serengil/deepf…

🧵 1/2 If you’ve ever spent hours stitching together ArcFace, VGGFace, age‑prediction nets, and a dozen other repos just to get a single face pipeline working, you’ll love what DeepFace just released.

2/2 It’s a 4‑billion‑parameter model that hits sub‑50 ms inference on a single A100 while staying under $0.001 per token – a sweet spot for real‑time apps. huggingface.co/openbmb/MiniCP…

🧵 1/2 Running LLMs in production feels like walking a tightrope: latency spikes and cloud bills can turn a promising feature into a cost nightmare. That’s why a growing number of teams are swapping hype‑driven giants for MiniCPM‑V‑4.6.

2/2 The repo already has 521 stars and 81 forks, so the community is testing it in the wild. github.com/knostic/OpenAnt

🧵 1/2 Tired of drowning in false‑positive alerts while hunting bugs? OpenAnt, the new open‑source LLM scanner, promises to cut the noise. It’s a Python‑based, multi‑provider agent that talks to any model you like, then runs a verified vulnerability workflow.

2/2 It’s the first open‑source suite that treats an agent like a software component you can unit‑test. github.com/claw-eval/claw

🧵 1/2 If you’re sick of ad‑hoc agent tests that never line up, meet claw‑eval. The repo ships 300 human‑verified tasks covering terminal automation, design tooling, and code generation—all in one place.

2/2 Then I found XcodeBuildMCP – a TypeScript‑based toolkit that ships a built‑in MCP server and a CLI, turning Xcode into a first‑class agent endpoint. No more brittle scripts, just a clean, structured interface. github.com/cameroncooke/X…

🧵 1/2 Every time I need to script Xcode builds for an iOS/macOS agent, I end up writing custom shell hacks and juggling config files.