An LLM browser harness in 600 lines, plus three more tools doing one job well

The Lens

Open Design is a local-first tool that turns your AI coding agent into a design studio. Point it at Claude Code, Cursor, Codex, or any of twenty-plus agents it supports, and it generates UI prototypes, dashboards, slide decks, images, and even video, all on your own machine, with no proprietary cloud in the middle. It's Apache 2.0 and positions itself as the open answer to tools like Vercel's v0 and Figma Make. The recent work has pushed it well past a simple prototype generator. It now ships with 150-plus built-in design systems, so outputs can auto-adopt a specific brand's tokens and look, plus a library of prebuilt skills for common jobs and a growing plugin marketplace. Video comes through HeyGen. Under the hood it's a Next.js and Electron app you can run as a desktop install or deploy with Docker. Setup is moderate, Node and pnpm, more moving parts than a single CLI but not painful. You bring your own model keys, and the media features lean on outside services like HeyGen that carry their own costs. It's genuinely useful if you want AI-generated design work without handing your prompts and outputs to a hosted platform. Solo builders and small teams experimenting with agent-driven design are the sweet spot. The catch is maturity. This is still a 0.x project moving fast, with a broadening scope and a healthy pile of open issues. The brand-design-system layer is brand new. If you need something stable and supported today, v0, Framer AI, or Lovable will feel more finished, at a subscription cost. If you want local-first and open, and you can tolerate the rough edges, Open Design is one of the few real options.

The Lens

Floci is a drop-in replacement for LocalStack that's actually free. This matters because LocalStack is sunsetting its free community tier in March 2026. Floci starts in ~24ms, uses ~13 MiB of memory, and the Docker image is ~90 MB vs LocalStack's ~1 GB. It runs on port 4566 (the same port LocalStack uses), so switching requires zero code changes. Built in Java with Quarkus, compiled to a native binary via GraalVM. MIT licensed. Already supports CloudFormation, Step Functions, DynamoDB Streams, Kinesis, Cognito, KMS, Secrets Manager, EventBridge, CloudWatch, API Gateway v2, and more. The catch: it's new and playing catch-up on AWS service coverage. LocalStack has years of edge cases handled. If you use an obscure AWS service, Floci might not emulate it yet. And 'no auth token required' means no telemetry, but also no commercial support if things break.

The Lens

whatcable answers a question macOS hides: what can this USB-C cable actually do? Plug a charger or peripheral into a Mac, and the menu bar tells you the cable's real specs (USB 2.0 vs 5/10/20/40/80 Gbps), its power rating (3A or 5A up to 60W/100W/240W), and why charging might be slower than expected. MIT licensed, free. The data already exists in macOS via IOKit. whatcable surfaces it. No private APIs, no kernel extensions, no daemons. Install via Homebrew (`brew install --cask whatcable`) or grab the signed and notarized .app from GitHub Releases. Apple Silicon only, macOS 14 or later. The "why is my Mac charging slowly" diagnostic is the killer feature. It tells you whether the cable, the charger, or the Mac itself is the bottleneck. This is a niche utility. If you have a drawer of identical-looking USB-C cables and one of them is silently a USB 2.0 charge-only cable, you have already needed this. Free, no paid tier, single-developer maintenance. The catch: Apple Silicon only. Intel Macs use older Thunderbolt controllers that do not expose the PD state and cable e-marker data the app reads.

The Lens

Browser Harness gives LLMs raw access to Chrome through a single WebSocket connection. No abstraction layer, no pre-built recipes, just direct CDP (Chrome DevTools Protocol) control. When the agent encounters something it cannot do, it writes new helper functions mid-task. Self-healing browser automation. The entire codebase is under 600 lines of Python. Connect to Chrome with remote debugging enabled, and your agent can navigate, click, fill forms, extract data, and extend its own capabilities on the fly. From the same team that built the browser-use framework, this is the stripped-down version for agents that need complete freedom. Developers building AI agents that interact with websites: this is the thinnest possible layer between your LLM and a real browser. The free tier at cloud.browser-use.com gives you 3 concurrent remote browsers for testing without managing Chrome instances. The catch: "complete freedom" means no guardrails. Your agent can navigate anywhere, click anything, submit forms. You need your own safety layer if you are pointing this at production accounts.