From Product Hunt
A Claude Code companion, a screen-level AI assistant, and a developer SDK that unifies 30+ storage backends.
CodeBreak — Claude Code Companion
Pixel-art character walks your screen while Claude Code runs, visible across every app. Celebrates on completion, panics when it needs input, sulks on errors. Four characters, eight sound packs. $7 one-time, macOS only.
Ludr AI — Screen AI That Lives Next to Your Work
Press a hotkey, draw a box around anything on screen, get an instant AI explanation. Supports GPT-4o and Gemini, inline text editing in Slack/Mail/WhatsApp, voice input, and chat threads. All data stays local. macOS and Windows.
Files SDK — One API for 30+ Storage Backends
Unified storage SDK for S3, R2, GCS, Azure, MinIO, Google Drive, Dropbox, and more. MIT licensed, web-standards I/O, escape hatch to native clients. Runs on Node, Bun, Workers, Vercel.
From Reddit
Anthropic overtakes OpenAI for the first time, Cerebras goes public with a bang, and GPT-5.5 runs autonomous science experiments.
For the first time in AI history, more US businesses are paying for Anthropic's Claude than OpenAI's ChatGPT. Anthropic's adoption rose to 34.4% vs OpenAI's 32.3% per the Ramp AI Index. Annualized revenue crossed $30B, surpassing OpenAI's ~$25B.
Claude Code is the engine: $2.5B+ annualized revenue, the fastest-growing product in Anthropic's history. Over 1,000 enterprise customers now spend $1M+/year. However, VentureBeat warns the lead may be fragile due to escalating compute costs and token-based pricing pressures.
GPT-5.5 autonomously spent 150+ hours improving protein folding models
Researcher Chris Hayduk reports GPT-5.5 running autonomously for over 150 hours optimizing protein folding prediction models, iterating on approaches without human intervention. Extends OpenAI's earlier work with Ginkgo Bioworks where GPT-5 ran 36,000 experiments over six months.
Cerebras IPO raises $5.5B, stock pops 108% on day one
AI chipmaker Cerebras went public at $66B valuation in the biggest tech IPO of 2026. CFO confirmed they're running GPT-5.4 and GPT-5.5 internally on their wafer-scale chips, with public inference coming soon. Backed by a $20B+ multi-year deal with OpenAI.
Figure AI robots sort 100K packages in 81 hours, fully autonomous
Three Figure 03 humanoid robots processed 101,391 packages over 81 continuous hours using Helix-02 onboard AI. Started as an 8-hour livestream challenge after a skeptic demanded proof, then just kept going. Zero human teleoperation.
M5 vs DGX Spark vs Strix Halo vs RTX 6000 — comprehensive benchmarks
All hardware in one room with standardized tests over 3 days. RTX 6000 leads at ~1,800 GB/s bandwidth vs M5 ~600 vs Spark ~256. Complete results published to a public repo. The practical takeaways match what the spec sheets suggest.
Quick hits from the local inference world.
Malta becomes first country to give all citizens free ChatGPT Plus
12 months of Plus access after completing a University of Malta AI literacy course. First national-level AI access program.
Qwopus3.5-9B-Coder — dense 9B model for agentic coding
Optimized for tool calling and agentic workflows. Runs at 8-bit on 16GB devices including Mac mini. Good sweet spot for local coding agents.
llama.cpp MTP prompt processing speedup
New PR avoids copying logits during prompt decode in MTP mode. Significant prompt processing improvement — time to update.
From Reddit
A local AI companion project, a practical 2D-to-3D pipeline, and the right way to prompt Klein's Qwen text encoder.
NeuralCompanion — open-source local-first AI companion
Desktop app combining realtime voice chat, local LLMs, TTS/STT, image generation, and a modular addon system. Supports avatar engines like VSeeFace and VAM. Hackable, API-friendly, runs on your own hardware.
Local pipeline: SDXL characters to game-ready 3D assets on 10GB VRAM
Custom automated workflow that takes SDXL character concept sheets and produces clean 3D meshes locally, without cloud APIs or heavy ComfyUI 3D nodes. Full engineering breakdown of the multi-stage pipeline.
How to actually prompt Flux.2 Klein with the Qwen text encoder
Klein 9B uses Qwen3 8B as text encoder with chat template wrapping. Your prompt is read like assistant instructions, not CLIP tags. Structured natural language in sections (subject, attire, expression, pose, background) beats tag-style prompting.
From Reddit
Arr stack gets a profile sync tool, a Swift clipboard manager with AI, and self-hosted replacements for vendor-locked hardware.
Profilarr v2 — sync quality profiles to your arr stack
Syncs quality profiles, custom formats, and media management settings from shared config databases (like Dictionarry) into Radarr/Sonarr. v2 adds multiple database support, import/export, and a web UI. Huge quality-of-life upgrade for arr users.
CopyFix — clipboard manager with LLM writing tools (macOS)
Native Swift clipboard manager with built-in AI writing tools. 8MB download, Apple Silicon native. Free tier with 10 AI requests/hour, or bring your own API key for unlimited use.
BLE Scale Sync — self-hosted smart scale bridge
Reads weight and impedance from 24+ BLE scales (Xiaomi, Renpho, Eufy, Beurer), calculates body composition locally, pushes to Garmin Connect, Home Assistant, and InfluxDB. No vendor cloud, no account needed.
Self-hosted maps and location history — complete Google Maps replacement
Full setup guide using Synology NAS, Traccar, Fedora, and Organic Maps. Replaces Google Maps bookmarks, saved places, and location history.