In partnership with

 
BUILDER HUSTLE
AI  ·  TECH  ·  STARTUPS
 
In This Issue
🔐 Perplexity Just Open-Sourced a Security Scanner for AI Developer Machines
🧠 A 26-Year-Old Built a YouTube Factory With Claude Code and Obsidian
⚡ Grok V9 Finished Training on 1.5T Parameters. 2 to 3 Weeks to Public Release.
👁️ Claude Mythos 1 Leaked Inside Claude Code. Adaptive Thinking Spotted.
🗣️ MIT-Licensed Talking Head Model Just Dropped. Production-Grade, Free to Use.
🐾 A Chinese Startup Says Its AI Collar Translates Pet Sounds With 95% Accuracy
🔐 PERPLEXITY · SECURITY · OPEN SOURCE
Perplexity Open-Sourced a Security Scanner Built Specifically for AI Developer Machines
Perplexity released Bumblebee, a read-only security scanner for macOS and Linux that checks developer machines for risky packages, suspicious browser extensions, and misconfigured AI tool setups including Claude Code and Cursor. It connects to supply-chain attack monitoring and was independently verified as clean with no backdoors or telemetry.
📌 Read-only scanner. It detects and reports only, it does not modify anything on your machine.
⚠️ Specifically flags Claude Code and Cursor configs. If your AI tools are misconfigured and leaking credentials, this catches it.
🔭 Supply-chain attacks on dev tools are rising fast. A scanner built for the AI tooling layer is overdue. This is the first serious one.
View on GitHub →

Are you running your business on incomplete numbers?

Most small business owners have financials, but few have financial clarity. There's a real difference between books that are technically up to date and books that actually tell you what's going on in your business right now. When accounting is reactive — updated when there's time, reviewed at tax season — you lose visibility exactly when you need it most. You can't tell which clients are truly profitable. You can't spot a cash flow gap before it becomes a crisis. BELAY's outsourced accounting team changes that.

Download the Free Guide

Get the free guide now →
🧠 CONTENT AUTOMATION · CLAUDE CODE · OBSIDIAN
A 26-Year-Old Built a Zero-Cost YouTube Factory by Giving Claude Code a Persistent Second Brain
A creator wired Claude Code directly into Obsidian as a persistent knowledge graph. Every piece of niche research, every script idea, every past video's performance data lives in Obsidian. Claude Code reads from it on every run, so each new video brief is built on full context, not a blank conversation.
📌 Obsidian acts as the long-term memory layer. Claude Code reads the graph on every run. No context is ever lost between sessions.
⚡ Zero external costs beyond a Claude subscription. Everything runs locally. No third-party APIs, no automation platforms, no extra spend.
🔭 Persistent memory architectures are the next leap in AI content workflows. A model with full context of your niche produces entirely different output than one starting cold.
See the full breakdown →
⚡ MODEL RELEASES · GROK · CLAUDE MYTHOS
Grok V9 Trained on Real Developer Sessions. Claude Mythos 1 Leaked Inside Claude Code. Two Big Moves in One Week.
Grok V9-Medium finished training this week according to TestingCatalog. It jumps from 0.5T parameters (V8-Small) to 1.5T and was trained heavily on real Cursor developer sessions, giving it grounded exposure to how actual engineers write and debug code. Fine-tuning and RL runs are next. Public release expected in 2 to 3 weeks. xAI is positioning this as a significant leap for difficult coding and agentic tasks. Separately, chetaslua spotted "claude-mythos-1-preview" appearing inside Claude Code and Claude Security UI strings. The model includes adaptive thinking, meaning it dynamically adjusts its reasoning depth per task rather than using a fixed chain-of-thought. New plan tiers also appeared alongside the leak. Anthropic has not officially confirmed Mythos 1 for general availability yet, but the strings are in the live product.
📌 Grok V9 trained on real Cursor sessions, not synthetic data. That's a meaningful difference for models targeting developer workflows.
⚠️ Claude Mythos 1 with adaptive thinking is unconfirmed for general public release. The UI strings are real but the rollout timeline is unknown.
🔭 Adaptive thinking means the model spends more compute on hard problems and less on easy ones. That changes the cost-per-task math significantly.
See the full breakdown →
🗣️ OPEN SOURCE · VIDEO AI · HUGGING FACE
A Potentially SOTA Talking Head Model Just Dropped on Hugging Face. MIT License. Free for Commercial Use.
Victor Mustar, Head of Product at Hugging Face, showcased LongCat-Video-Avatar-1.5 by meituan-longcat this week. It's an open-source talking-head model with strong lip-sync and identity preservation, described by Mustar as potentially SOTA. The model and a live demo Space are both publicly available on Hugging Face. MIT licensed, meaning it's free for commercial products with no restrictions.
📌 MIT license means you can build commercial products on top of this right now. No royalties, no usage restrictions, no approval needed.
⚡ Strong lip-sync and identity preservation are the two hardest problems in talking-head video. Mustar specifically called both out as working well.
🔭 Talking-head agents that can hold a consistent visual identity across sessions are the next layer of AI tutoring and sales tools. This model makes that buildable today.
Try the live demo →
🐾 CONSUMER AI · PET TECH · HYPE CHECK
A Chinese Startup Claims Its AI Collar Translates Pet Sounds Into Sentences. Replies Are Skeptical.
Min Choi shared a Chinese startup's AI pet collar claiming 95% accuracy translating barks and meows into full sentences in 1.2 seconds. The underlying model, "Pet-LLM," was trained on 1.5M+ real-world audio samples and 3,200 hours of annotated video. A demo circulated widely and the product got significant traction on X.
📌 1.5M+ audio samples and 3,200 hours of annotated video is a serious dataset for a niche product. The training effort here is not trivial.
⚠️ The 95% figure almost certainly applies to trained behavior categories like hunger and play. Open-ended understanding of animal intent is not what this does.
🔭 Purpose-built small models trained on domain-specific data shipping in consumer hardware is the actual trend here. Pet-LLM is an early example of that pattern.
See the original post →

You're reading about Grok, Claude, and AI agents all day - now hire one. Viktor lives in Slack, connects to 3,000+ tools, and quietly ships the work while you sleep.

Your next great hire lives in Slack.

Viktor is an AI coworker that connects to your tools and ships real work. Ask Viktor to pull a report, build a client dashboard, or source 200 leads matching your ICP. Most teams hand over half their ops within a week.

Add Viktor to Slack for free.

Get $100 in Viktor credits →

Today's Top Launches

Products To Watch Out Today

🏆 AltPage.ai

Marketing · SEO · Artificial Intelligence

Steal competitor brand traffic with alternative pages. It auto-generates comparison and alternative landing pages targeting searches like "best alternative to [competitor]" - a sneaky but smart way to poach high-intent traffic.

🔍 LLM SEO Monitor

Marketing · SEO · Artificial Intelligence

Monitor what ChatGPT, Google Gemini, and Claude recommend. If you want to know whether your brand shows up when someone asks an AI for a product recommendation, this tool tracks exactly that - essential for AI-era SEO.

🧩 Kibo UI

Open Source · Developer Tools · GitHub

Advanced, open-source components for shadcn/ui. Think of it as a supercharged extension pack, more complex, production-ready components built on top of the shadcn system for devs who need more than the basics.

See All Launches →
 

Stay sharp,
Builder Hustle

Keep Reading