From Product Hunt

Heavy agent infrastructure day. Multiple products building the plumbing for AI agents to act in the real world — payments, identity, browsing, and deployment.

Needle 2.0 — Vibe-Automate Your Workflows Describe what needs automating in plain language, and an AI agent builds, tests, and ships the workflow. Connects to Google Drive, Notion, Slack. Marketplace to publish and earn from workflows. AI-first competitor to Make/n8n.
▲ 440 · needle.app
HTML Pub — Publish AI-Generated HTML via MCP Deploy live websites without leaving your AI chat. Works with Claude, ChatGPT, Cursor, Windsurf via MCP. Custom domains, forms, analytics, heatmaps. From the Leadpages CEO. Free tier: 5 pages.
▲ 363 · htmlpub.com
Prava — Payments SDK for AI Agents Visa-partnered payments infrastructure for agents. Tokenized, PCI-compliant agentic payments with spending limits, passkey approvals, and guardrails. 4 lines of code to integrate. Already live powering shopping agents.
▲ 280 · prava.space
Agent Browser — Token-Efficient AI Browsing Renders web pages as ASCII wireframes with numbered elements instead of screenshots. Agent sees "[12]Sign Up" instead of a 1280x720 image. 3-10x fewer tokens per page. Open source, works as MCP server for Claude Desktop/Cursor.
▲ 128 · agent-browser.io
Clawcard — Give AI Agents a Full Identity Provision a credit card (via Privacy.com), email inbox, and SMS phone number for AI agents with one API key. Agents can make purchases, receive verification codes, and manage accounts autonomously.
▲ 112 · clawcard.sh
Airpoint — Touchless Computer Control Control any computer with hand gestures using just your webcam. Real-time hand tracking for cursor control, shortcuts, and scrolling. SDK for apps/games. Also enables OpenClaw agents to control computers physically.
▲ 106 · airpoint.app
From Reddit

The Anthropic-Pentagon saga intensifies with the clearest explanation yet of why DoD labeled Claude a supply chain risk. Meanwhile, Meta's next-gen model stumbles, and Google's Gemini starts doing errands on your phone.

Bernie Sanders Introduces AI Data Center Moratorium Sanders officially introduced legislation calling for a 6-month moratorium on all new AI data center construction, citing existential threat and the need for democratic debate. Massive community reaction (2,500+ upvotes, 761 comments). Top comment: "China is not going to stop."
Meta Delays "Avocado" Model to at Least May Update: Meta's next-gen model failed to match Gemini 3.0, OpenAI, and Anthropic on internal reasoning, coding, and writing tests. Meta AI leaders reportedly discussed temporarily licensing Gemini to power Meta's AI products. Significant setback for their AI ambitions.
Gemini Task Automation Rolls Out on Android Gemini's multi-step task automation is live on Samsung Galaxy S26 and rolling out to Pixel 10. Can order Uber rides, food delivery, and groceries autonomously. Runs apps in a secure virtual window and confirms before completing actions. US and Korea first.
Meta Reveals Four Custom MTIA Inference Chips MTIA 300, 400, 450, and 500 on a 6-month cadence. MTIA 500: 30 PFLOPS MX4, 27.6 TB/s HBM bandwidth, 1700W TDP. Inference-first design (opposite of Nvidia's training-first). PyTorch-native with vLLM support. 25x compute increase from 300 to 500.
OmniCoder-9B — Agentic Coding Model 9B coding agent fine-tuned on 425K+ agentic trajectories from Claude Opus 4.6, GPT-5.4, and Gemini 3.1 Pro. Built on Qwen3.5-9B's hybrid architecture. Shows strong error recovery, LSP diagnostics, and proper edit diffs. Community reports it one-shots tasks that base Qwen3.5-9B fails at.
From Reddit

FLUX 2 gets a speed boost and Nvidia's RTX nodes prove their worth in an upscaling showdown.

FLUX.2 Klein 9B Gets KV-Cache for 2x Faster Multi-Reference Editing Black Forest Labs released an optimized FLUX.2 Klein 9B variant with KV-cache support. Caches key-value pairs from reference images during first denoising step, eliminating redundant computation for up to 2x faster editing workflows. FP8 quantized weights built with Nvidia.
Nvidia RTX Super Resolution vs SeedVR2 Upscaling Head-to-head comparison of real-time Nvidia RTX upscaling vs SeedVR2 (6 sec) on FLUX Klein 9B images. Community consensus with 710 upvotes: Nvidia looks more natural with no artificial smoothing. RTX is interpolation-based, SeedVR2 is generative — different strengths for different needs.
From Reddit

A strong crop of macOS utilities this week — local voice cloning, file conversion by renaming, and several menu bar tools.

Morpholder — Convert Files by Renaming in Finder Menu bar app that monitors folders and converts files when you change extensions. PNG to ICO, HEIC to JPG, MOV to GIF. Smart suffixes like _nobg for AI background removal and _text for OCR. 100% local, Apple Silicon optimized. $14.99 lifetime.
Speaklone — Local Voice Cloning on Apple Silicon Native macOS TTS and voice cloning via MLX (Qwen3-TTS). Clone any voice from 3 seconds of audio. Voice design from text descriptions. System-wide dictation, local API server for integration. 10 languages, completely offline. One-time purchase.
CodePeek+ — Syntax-Highlighted Quick Look for Code Free Quick Look extension. 40+ languages, Markdown rendering, Jupyter notebook preview, custom themes (VS Code Dark+, Dracula, Nord). Just press Space on a code file in Finder.
Chunk v2 — Menubar Time Blocking Syncs with Apple/Google/Outlook calendars. Fullscreen alerts when blocks end, countdown timer during tasks. Loved by the ADHD community. $14.39 one-time.
Sonexis — Visual Audio Effects Chain for macOS Drag-and-drop blocks to build custom system-wide audio processing chains. Serial or parallel paths, real-time effects on all audio. Free.
Usage4Claude — Claude Quota Monitor in Your Menu Bar Free open-source app showing 5-hour, 7-day, extra, Opus, and Sonnet quotas with color-coded indicators. v2.5 adds built-in browser login. Multi-account support.