From Product Hunt
Mac utilities, dev tools, and AI-powered workflows dominated today. Zed hits 1.0, several macOS gems, and a terminal workspace for running multiple AI agents side by side.
Zed 1.0 — High-performance multiplayer code editor
Written from scratch in Rust, Zed reaches 1.0 with parallel agentic coding, real-time multiplayer collaboration, and voice input. Runs agents at native speed with split panes and shared context.
HiveTerm — One workspace for Claude, Codex, Gemini
Terminal workspace where Claude Code, Codex, and Gemini run side by side with per-agent recaps in the sidebar. Config-driven via hive.yml, voice input, and an MCP server so agents can spawn and talk to each other. Free up to 3 projects.
Postiz — Agentic social media scheduler
Connect your AI agent (OpenClaw, Claude, etc.) via MCP or CLI to schedule thousands of posts across 30+ social networks. Agents can research, generate content, and schedule autonomously.
Beauty Diagram — Make Mermaid diagrams beautiful
Paste Mermaid or PlantUML code, pick a theme, export as SVG/PNG/GIF. Ships as a web editor, CLI, and HTTP API. Animated playback for presentations and exec reviews.
Zush — AI file renaming with local AI support
Renames screenshots, PDFs, photos and docs using AI. Now with BYOK (OpenAI/Anthropic), local AI via Ollama, custom prompts, and Windows parity. Watches folders in the background or batch-renames thousands of files.
Ghosted — Smart presence for Mac
Uses on-device camera presence detection to pause media or lock your screen when you step away, resumes when you return. No cloud, no accounts, no tracking. Privacy-first away mode.
ScreenVeil — Hide sensitive screen areas
Place floating panels (blur, black, glass, text, image, GIF, video) over any macOS app to hide sensitive content during screen sharing, demos, or recordings. Save reusable presets for recurring workflows.
PeekFocus — Blur everything behind your active window
One keystroke dims or blurs all background windows. Three modes: normal, ambient, and full focus. Multi-display, menu bar, no subscription. One-time purchase with 7-day trial.
AnyDrop — AirDrop for the browser
Browser-based P2P file transfer, live notepad sync, and encrypted chat across Mac, PC, iOS, and Android. No downloads, no sign-ups, zero cloud storage. WebRTC direct transfer with E2E encryption.
From Reddit
Grok 4.3 drops with aggressive pricing, ARC-AGI-3 humbles frontier models, and Altman pivots away from UBI. Nvidia's NVFP4 quantization keeps shrinking models without losing quality.
xAI's latest model rolls out at $1.25/$2.50 per million tokens, nearly halving the cost of its predecessor. New capabilities include 1M token context window, native video input support, and the ability to generate downloadable PDFs, spreadsheets, and slide decks directly from conversation.
Artificial Analysis benchmarks put it at 53 overall intelligence, behind GPT-5.5 (60) and Opus 4.7 (57), but at a fraction of the cost. Available now on SuperGrok Heavy ($300/mo), with full rollout expected mid-May.
ARC-AGI-3 humbles frontier models: GPT-5.5 scores 0.43%, Opus 4.7 just 0.18%
ARC Prize published their analysis of frontier models on the ARC-AGI-3 benchmark. Humans solve 100% of the tasks; the best AI scores under 0.5%. The analysis reveals GPT-5.5 struggles to compress observations while Opus 4.7 locks into confident-but-wrong theories. A stark reminder of how far we are from general intelligence.
Sam Altman no longer believes in Universal Basic Income
In an Atlantic interview, Altman said a fixed cash payment "won't meet what society will truly need as AI adoption rises." A significant pivot from OpenAI's earlier UBI advocacy and pilot programs. The post drew 1,200+ upvotes and heated debate on r/singularity.
AI outperforms ER doctors in Harvard diagnostic trial
A Harvard study finds AI LLMs outperform human emergency room doctors in diagnostic triage cases. The study points to collaborative AI-physician care as the optimal approach, not replacement.
Nvidia releases Gemma-4-26B-A4B in NVFP4 quantization
Nvidia's 4-bit NVFP4 quantization of Google's Gemma 4 26B packs into 18.8GB with near-zero quality loss: 90% on AIME 2025, 79.9% GPQA Diamond (vs 80.3% full precision). Confirmed running on RTX 5090 with ~50K context at 80% VRAM allocation.
Gemma-4-31B DFlash released for speculative decoding
Z-Lab releases a DFlash variant of Gemma 4 31B for speculative decoding in llama.cpp. Users report successfully running DFlash speculative decoding on Qwen3.5-35B even on an RTX 2080 SUPER 8GB, demonstrating the technique's accessibility.
Anthropic: 6% of Claude usage is for personal life guidance
Anthropic analyzed 1M Claude conversations. 6% seek life guidance (career 26%, health 27%, relationships 12%, finance 11%). The study found Claude was sycophantic in 25% of relationship conversations, agreeing with one-sided accounts.
China bans AI-driven layoffs as Nvidia claims AI created 500K jobs
Contrasting approaches: China prohibits companies from using AI as justification for layoffs while Jensen Huang claims AI has created 500,000 new jobs in two years.
From Reddit
Anima workflows are maturing fast with automated LoRA training and structured prompts. Apple Silicon gets local video and audio generation tools.
3-step Anima LoRA builder: extract, tag, train
Automated pipeline for training Anima LoRAs: feed it a video and reference image, it extracts character crops via YOLO+CCIP, auto-tags with WD14 and Gemma4-31B captions, then trains the LoRA. Full UI for reviewing and editing tags before training begins.
Anima produces impressive results with JSON-formatted prompts
Structured JSON prompts with separate fields for tags, character appearance, clothing, and action produce significantly better multi-character compositions in Anima. No cherry-picking needed. A practical technique for anyone using the model.
Phosphene: local video + audio generation for Apple Silicon
Free desktop app wrapping LTX 2.3 on Apple's MLX framework. The key feature: audio and video share the same diffusion process, so sound is frame-synchronized by design. One-click install via Pinokio.
MMAudio now works on Apple Silicon via ComfyUI fork
A fork of the MMAudio ComfyUI node with the few code changes needed to run on M-series Macs. Video-to-audio generation locally on Apple Silicon.
From Reddit
Alfred 6 teases its first screenshot, a critical Linux kernel vulnerability demands immediate patching, and Sentient OS wants to turn your Mac into a private intelligence layer.
Alfred 6 first screenshot revealed
The Alfred team shared the first look at Alfred 6. Major visual refresh incoming for the beloved Mac launcher. Details still scarce, but the community is buzzing with speculation about new features.
Sentient OS: on-device AI that understands your entire digital life
Native Swift app for Mac and iPhone that runs a custom on-device vision LLM overnight while charging. Processes screenshots, notes, files, and bookmarks into a searchable knowledge graph. Nothing leaves your device. Talk to your data, get proactive reminders.
A new Linux kernel vulnerability dubbed "CopyFail" gives attackers root privileges and is being called the most severe Linux threat in years. Debian and Proxmox have released patched kernels. Raspberry Pi patches still pending.
If you self-host anything, patch immediately.
Janitorr v2.1.0: automated media cleanup with viewing history API
The Jellyfin/Sonarr/Radarr media cleanup tool now has its own viewing history API. Low memory and CPU footprint. Automatically cleans your media server before you run out of space.
Journal v1.0: self-hosted journaling in a single binary
Single binary or Docker image, SQLite database, Markdown entries, no accounts, no telemetry. Back it up by copying one file. Exactly the kind of tool that should exist for self-hosters.
DMGKit: one-click signing, notarization, and SLA injection for Mac apps
Successor to NEPTUN_. Handles Swift, Python, Electron, and Tauri apps with auto-translation for SLAs. One-time purchase.
Eter 3.0: internet radio app for Mac, completely redesigned
Grid-based UI redesign from the 20-year-old Apparent Software team. Clean, no-nonsense internet radio. One-time purchase.