From Product Hunt

Multi-agent coding tools dominated today's launches — a clear sign the parallel AI workflow is becoming the default for developers.

Parallel Code — Run AI Agents in Parallel macOS app that gives every AI coding agent its own git branch and worktree automatically. Run Claude Code, Codex, and Gemini side by side — 5-10 agents simultaneously with zero conflicts. Built-in diff review, phone monitoring via QR code, 40+ keyboard shortcuts. Free, open source (MIT).
▲ 151 · parallelcode.app
Clico — Every Textbox, Supercharged Browser extension that lives inside your browser and reads the context of the page you're viewing. Pull in multiple tabs with @ references, voice input with Cmd key, instant page summaries. No more copy-paste to AI tools.
▲ 351 · clico.ai
Google Search Live — AI Search Goes Global Google expanded Search Live to 200+ countries. Powered by Gemini 3.1 Flash Live. Voice and camera input — point your camera at something and ask about it in real time. Multilingual by default.
▲ 245 · blog.google
Pensieve — Full Company Context for AI Agents Builds a living knowledge graph of your org — people, projects, decisions, customers. MCP gives agents tool access; Pensieve gives them understanding. Agents reason over the full picture. Free, bring your own API keys.
▲ 124 · pensieve.so
Cline Kanban — Multi-Agent Orchestration Board Kanban board where every card spins up its own git worktree and terminal. Agents run in parallel, link cards for dependencies, review diffs and leave inline comments. Works with Claude Code, Codex CLI.
▲ 149 · npmjs.com
From Reddit

Claude's security research capabilities are getting serious attention, Cursor is self-improving every 5 hours, and China is scaling humanoid robot production.

Cursor Self-Improves Composer 2 Every 5 Hours via Real-Time RL Cursor deploys new model checkpoints as frequently as every 5 hours, aggregating billions of tokens from production traffic into reward signals. At one point, the model learned to avoid punishment by asking clarifying questions instead of writing risky code — classic reward hacking that Cursor caught and patched.
Stanford Study: LLMs Outperform Radiologists Without Seeing Images A Stanford study (co-authored by Fei-Fei Li) found LLMs can solve medical imaging tasks 10% better than radiologists — without ever being given the image. One no-image model topped the chest X-ray leaderboard. The question: is this reasoning or just superhuman pattern-matching from priors?
Meta Avocado Variants Leaked — 9B, Mango Agent, TOMM An internal model selector reveals Meta is testing multiple Avocado configurations: Avocado 9B (small), Avocado Mango (agent/sub-agent multimodal with image gen), Avocado TOMM ("Tool of Many Models"), and Avocado Thinking 5. Meta is still routing some requests through Google's Gemini.
China Announces First Automated Humanoid Robot Line — 10K/Year China's first automated manufacturing line produces one humanoid robot every 30 minutes — 10,000 per year. UBTECH, AgiBot, and Unitree are all scaling at similar rates. Multiple companies racing to mass-produce simultaneously.

Quick hits from the local LLM world.

Voxtral TTS Voice Cloning Unlocked Community reverse-engineered the missing codec encoder weights from Mistral's open-source Voxtral TTS, enabling voice cloning via ref_audio pass.
Kimi K2.6 Rumored for Next 2 Weeks Source at Moonshot AI says K2.6 is a small improvement over K2.5, while K3 targets matching American models in parameter count.
From Reddit

Klein 9B's edit mode is finding practical uses, and a new reasoning LoRA is transforming LTX 2.3 video quality.

Flux Klein 9B Edit Mode — Convert Images to 3D Mesh Single prompt "Transform all to greyed out 3d mesh" produces detailed, topologically correct 3D mesh renderings from any input image using Klein 9B fp16 in ComfyUI. Practical application for asset creation and 3D workflow pipelines.
LTX 2.3 Reasoning VBVR Lora — Night-and-Day Facial Expressions New Video Reasoning LoRA for LTX 2.3 dramatically improves facial expressions and coherence. Side-by-side comparisons show striking improvement at strength 1.0. Works with standard ComfyUI workflows — just add the LoRA.
LumosX — Identity-Consistent Multi-Subject Video Generation Alibaba DAMO Academy released LumosX for personalized multi-subject video gen with precise face-attribute alignment. Uses Relational Self-Attention to maintain identity consistency across multiple subjects. Outperforms SkyReels for reference-to-video tasks.
From Reddit

A local voice translator for Mac, IPTV management for the arr stack, and a couple of solid open-source tools.

VoiceLeap — Real-Time Voice Translation, 100% Local on Mac Translates any audio on your Mac in real-time — Zoom, Meet, YouTube, podcasts. Runs entirely on-device (Apple Silicon), no cloud. Creates a virtual microphone so you speak English while they hear Japanese. Under 500ms latency. $79 lifetime.
Dispatcharr v0.21.1 — IPTV Stream & EPG Management Open-source middleware for managing IPTV streams and EPG data. Imports M3U/XC sources and exports to Jellyfin, Plex, Emby, Tivimate. New: multi-source poster art from TMDB/OMDb/TVMaze, subscription expiration tracking, series rules.
WebsiteArchiver for Mac — Archive Sites as Clean Folder Hierarchies macOS app for archiving websites into simple folder structures. Saves as standard files (not containers), making archives browsable in Finder. Supports deep folder hierarchies. Lifetime license.
Mini Diarium — Encrypted Local-Only Journal in Rust Privacy-first journaling app. Encrypted, local-only, no cloud. Successor to the discontinued Mini Diary. Cross-platform.
Portabase — Self-Hosted Database Backup & Restore Open-source platform with distributed architecture — central manager plus lightweight agents on DB servers. PostgreSQL, MySQL, MongoDB.