From Product Hunt
Coding agent tools and macOS utilities lead today's launches.
Panorama — AI that finds your team's hidden workflows
Analyzes your workplace data to discover repetitive patterns and recommend AI workflows your team can automate. Integrates with Slack, Notion, and Google Workspace.
Shotwell — The screenshot editor for iPhone
Adds device frames, backgrounds, and shadows to iPhone screenshots natively on your phone. No more AirDrop-to-Mac roundtrip. Like CleanShot X but for iOS. Free.
Handle — Refine UI visually, feed changes to your coding agent
Open source Chrome extension + MCP that lets you point-and-click to tweak UI in the browser instead of re-prompting your agent. Works with Claude Code, Cursor, Codex.
CatBar — RevenueCat stats in your macOS menu bar
Native menu bar app showing MRR, subscribers, and transactions from RevenueCat at a glance. Real-time notifications, charts, and AI insights. Credentials in Keychain, no middleman server.
Guardian IDE — Control AI-generated code before it ships
Desktop release gate for AI-generated code. Local code review, policy enforcement, release approvals. Treats AI output as something needing governance.
From Reddit
Anthropic flips the revenue lead, GPT Image 2 leaks, and Gemma 4 is crushing benchmarks locally.
SemiAnalysis reports Anthropic has hit $25B ARR, overtaking OpenAI's ~$24B. Anthropic tripled from $9B to $25B in a single quarter — driven largely by Claude Code, which alone has $2.5B+ in run-rate revenue.
The revenue gap that was $6B in Anthropic's favor of OpenAI has now flipped entirely. Enterprise adoption accelerating.
GPT Image 2 Leaked on Arena — Massive Quality Jump
Three unnamed models appeared on Chatbot Arena under tape-themed codenames. Community consensus: a meaningful step above anything publicly available. Yellow filter fixed, text renders correctly, beats Nano Banana Pro on photorealism. Release expected within weeks.
Update: Gemma 4 31B Is Destroying Community Benchmarks
Following this week's release, community testing shows Gemma 4 31B outperforming GPT-5.2, Gemini 3 Pro, and Sonnet 4.6 at 180x less cost than Opus 4.6. The 26B MoE variant is the new king for 16GB VRAM setups. Per-layer embeddings architecture is the key innovation.
Real-Time Local AI (Audio + Video In, Voice Out) on M3 Pro
Gemma 4 E2B running real-time multimodal AI entirely on a Mac — camera input, voice output, multilingual. Essentially what OpenAI demoed years ago, now running locally on consumer hardware. Great for language learning.
Chinese AI Labs All Pausing Open-Source Releases at the Same Time
MiniMax M2.7, GLM-5.1, Qwen 3.6, MiMo-V2 Pro — all delayed open-source with identical "improving before release" messaging. Community suspects coordination. Pattern doesn't feel organic.
Society & AI — how AI is reshaping skills and writing.
"Everyone Now Sounds the Same" — How AI Is Changing College
CNN reports on AI homogenizing student writing and thinking patterns. Professors note students produce similar-sounding work regardless of whether they used AI directly.
11-Year Developer Realizes He Can't Debug Without AI
Developer shares how he found himself unable to methodically debug his own code without reaching for Claude first. Sparked discussion about AI dependency and skill atrophy.
From Reddit
ComfyUI ecosystem keeps evolving — new optimizations and asset management tools.
Accelerate Z-Image by 20-30% With This ComfyUI Node
Open-source custom node using Triton kernel fusion + W8A8 INT8 quantization to speed up Z-Image S3-DiT by 20-30% and save 3.5GB VRAM. Works on existing BF16 models with no extra downloads.
ComfyUI Assets Manager Gets Gallery, Fast Indexing, and Drag & Drop
Major update: integrated gallery for all outputs/inputs, lightning-fast indexing even with massive libraries, drag-and-drop back into workflows, smart filtering, and a pop-up viewer for hi-res inspection.
Z-Image + LTX 2.3 Achieving Cinematic Realism Beyond Proprietary Models
Workflow combining Z-Image Turbo for frames with LTX 2.3 img2vid produces more natural results than even Seedance 2.0. Z-Image Turbo remains the best overall model for style and speed.
From Reddit
Self-hosted hardware, dev tools, and open-source finds.
Nomad Mk3 — Pocket-Sized Offline Media Server on ESP32-S3
Fully self-hosted media server on an ESP32-S3. Creates its own WiFi and serves movies, shows, music, books, and files to any browser. Multiple simultaneous users, ultra low-power, open source. Uses an SD card for storage.
Mockphine — Local Mock API Server for macOS
Desktop app for running a local mock API server. Per-endpoint modes (mock/passthrough/disabled), Live View traffic inspector, delay simulation. One-time purchase. macOS + Windows.
Kreuzberg v4.7.0 — Document Intelligence for 248 Languages
Rust-core document intelligence library supporting Python, TypeScript, Go, Ruby, Java, C#, PHP, and more. New markdown rendering, HTML output, OpenWebUI integration, and code intelligence for 248 languages.
Kokoro TTS Voice Training Now With GPU/CUDA Support
Fork of KVoiceWalk that adds GPU acceleration and a GUI queue to Kokoro TTS training. Drops training time from ~26 hours on CPU to minutes on GPU.
Hacki — Open-Source Hacker News Client
Free, open-source Hacker News client with a clean interface for browsing stories, comments, and user profiles.