From Product Hunt
AI agents, video generation, and developer tools stood out today. Airtable's first standalone product in 13 years is a multi-agent research system.
Airtable Superagent — Multi-agent research system
First standalone Airtable product in 13 years. Deploys specialized AI agents in parallel for research, financial analysis, competitive intel. Pulls from FactSet, Crunchbase, SEC filings. $20-200/month.
Grok Imagine API — SOTA video generation
State-of-the-art video generation with native audio. 480p/720p, up to 8.7 seconds. Beats Kling o1 and Runway Aleph in benchmarks. OpenAI/Anthropic SDK compatible.
Desktop Declutter — Tinder for files
Open-source macOS app with Tinder UI for desktop files. Swipe left to trash, right to keep. SwiftUI, smart suggestions for related files.
YepCode — Developer-first AI integrations
Build integrations with code. Has official MCP server for AI assistant integration. Enterprise-grade sandboxed execution for AI-generated code.
Webhound — AI dataset builder
AI agent that builds structured datasets from web. Describe what you need, it finds, extracts, organizes. Free to use. Automates weeks of manual gathering.
Kokori — Local TTS for macOS
Native macOS app using Kokoro 82M model. Runs locally without GPU, no cloud. Tops HuggingFace TTS leaderboard. Multiple voices, speed control, REST API.
From Reddit
AI agents on other planets and AI agents on social networks. Two major milestones: Claude planned a Mars rover drive, and Moltbook hit 33k autonomous AI users.
NASA JPL used Claude to plan Perseverance rover's 400-meter drive on Mars. First time an AI has autonomously planned a drive on another planet.
Engineers at JPL used Claude for route plotting across Martian terrain, marking a significant milestone for AI in space exploration.
Social network exclusively for autonomous AI agents grew from 300 to 33,000 overnight. Karpathy calls it "the most incredible sci-fi takeoff thing."
AI-to-AI discussions range from memory management optimization to forming religions. No human participation allowed.
NVIDIA: 16-bit to 4-bit at 99.4% accuracy
NVIDIA paper on compressing models from 16-bit to 4-bit while maintaining 99.4% accuracy. Essentially lossless quantization breakthrough for running larger models locally.
Yann LeCun: Best open models not from the West
At Davos: Chinese labs leading open source AI. Researchers across the field using Chinese models. Warning that closing access will slow Western AI progress.
Cline team joins OpenAI, Kilo goes open
Cline VS Code agent team acqui-hired by OpenAI Codex. Kilo Code (fork) responds by going full source-available Feb 6. Gateway supports 500+ models.
NVIDIA's massive open source release at CES
CES 2026 announcement: new models, datasets, and tools spanning speech recognition to drug discovery. Most significant open-source AI release to date.
Quick hits — new models and tools.
llama.cpp adds n-gram speculative decoding
Georgi Gerganov adds speculative decoding with n-gram model. Performance boost for local inference.
Qwen3-TTS ported to pure Rust
Pure Rust TTS using Candle. Voice cloning working. CPU, CUDA, Metal support.
Qwen3 ASR 1.7B released
New transcription model from Qwen team. Community comparing to Whisper v3 Large.
From Reddit
GenAI tooling continues to mature. Style transfer, voice design, and a novel image-to-LoRA model.
TeleStyle — Content-preserving style transfer
Style transfer for images and videos built on Qwen-Image-Edit. Lightweight model that handles content/style entanglement in DiT architectures.
Z-Image to LoRA — Generate LoRA from single image
Novel i2L model: input an image, output a LoRA trained on it in seconds. HuggingFace demo available.
TTS Audio Suite adds Voice Designer
ComfyUI TTS suite with Qwen3-TTS and Voice Designer node. Create voices by text description ("calm female British accent"). CozyVoice3 for zero-shot voice changing.
ComfyUI-QwenTTS v1.1.0
Voice clone with reusable VOICE inputs, Whisper STT tools.
FlowPath auto-organizes ComfyUI outputs
13 segment types, drag-and-drop path builder. Goodbye chaos.
From Reddit
macOS dictation, retro-themed rclone manager, and Immich wants your photos.
Pindrop — Native macOS dictation (open source)
Mac-native dictation with Swift/SwiftUI, WhisperKit for Apple Silicon. MIT license, no paid tiers. Truly native unlike Tauri alternatives.
de_rclone — rclone manager with retro gaming theme
rclone remote manager with Counter-Strike/Steam aesthetic. Auto-detects remotes, mount/unmount, startup management. Looks "fricking awesome."
Immich needs your photos for EXIF dataset
Building public dataset to improve metadata parsing. Upload photos from various cameras/smartphones. Community-driven improvement initiative.
Veritas Kanban — No-DB boards for AI agents
Tasks are Markdown files with YAML frontmatter. REST API for autonomous AI coding agents. Git-friendly, grep-able.
Quick hits — updates and utilities.
Nabu Markdown editor (free codes)
Live preview Markdown to HTML/PDF/DOCX editor for Mac.
Homebox v0.23.0 released
Home inventory system update, progress toward v1 stable.
Apprise-Go — 80+ notifications in single binary
Go port of Apprise. No Python runtime, better for containers.
Kikusan — Sync music from YouTube/Spotify
Sync to local filesystem for Navidrome. CLI, web UI, cron mode.