OpenClaw (Clawdbot / Moltbot) Videos & Summaries

The House and its Mysterious Door

• The video recounts the story of Eleanor Wentworth inheriting an old Victorian mansion in Milbridge, Connecticut.
• She discovers a peculiar basement door that opens inward, unlike typical Victorian doors, and is covered in claw-like scratch marks on the outside.
• This door has been sealed since the disappearance of her great-uncle Nathaniel Wentworth in 1945.

Uncovering Dark Family Secrets

• Eleanor finds Nathaniel's journal, detailing his fear of something in the basement and strange experiments.
• Clara, Nathaniel's fiance, also kept a journal describing a trapped shadow creature born from grief and cunning.
• Historical records and local testimonies reveal the Wentworth family's involvement with an occult society and a pact made by great-grandfather Horus to imprison a darkness.

The Nature of the Darkness

• Nathaniel's experiments were an attempt to contain this darkness using alchemy and arcane rituals, making himself a vessel.
• The scratch marks are described as both warnings and invitations, an attempt by the creature to be remembered and freed.
• The entity is not just a physical being but a manifestation of the family's sin, sorrow, madness, and betrayal.

The Confrontation and Revelation

• Eleanor descends into the basement, finding symbols etched on the walls and a grate concealing a deep shaft.
• The scratching intensifies, and a voice whispers her name, revealing the creature is tied to her bloodline.
• The door opens inward not to keep people out, but to keep the entity contained, a seal strengthened by rituals and blood.

The Ongoing Curse

• The creature's hunger is for remembrance and identity, feeding on pain and the shattering of reality.
• Eleanor realizes the curse is passed through generations, making the family custodians of suffering.
• The choice remains between perpetuating the cycle of silence and suffering or facing potential destruction by opening the door wider.

AI Agents in Business

• AI agents are becoming a foundational tool, akin to word processors for personal computers or apps for mobile.
• A key application of agents is software development (coding agents), enhancing productivity for existing engineers and enabling "vibe coding" for prototypes.
• Another killer app is the "run your company better" agent, acting as a knowledge base, data analysis, and project management tool.
• Vercel's internal agent, "V," assists with customer support, sales, marketing content, and engineering blog posts.
• V has sub-agents, acting like a virtual employee team for specific tasks.

Building and Deploying Agents (Eve Framework)

• The "soul" of an agent is defined in a markdown file, similar to Open Call's soul.md, detailing its genesis and instructions.
• Eve provides a framework for building agents, analogous to Next.js or React for web development.
• Agents need a secure and efficient environment to run, with configurable tools, human approvals, and data access controls.
• Agents can "hibernate" when not in use, similar to serverless computing, for efficiency.
• Eve allows agents to be connected to various communication channels like Slack, WhatsApp, and iMessage.

Agent Capabilities and Future

• Agents can be skilled in specific tasks (e.g., content writing) and integrate external APIs or "plugins" for enhanced functionality.
• Proactive agents can perform tasks autonomously, like daily data analysis or social media monitoring, without explicit prompting.
• Connecting agents to external systems (e.g., Stripe, email) via event-driven triggers is crucial for automation.
• The future of business involves building agents before websites, serving as a company's core intelligence and trusted partner.
• Self-improvement through evaluations, tests, and feedback loops is key for agent evolution.
• The cost of intelligence is decreasing, and models are becoming faster, making agents more accessible and effective.
• Model agnosticism allows users to choose the best models for specific tasks and avoid vendor lock-in.

OpenAI Setup and Gemini Integration

• Gemini API key configured and Gemini onboarded.
• Default model overwritten with Gemini 3.1 Pro.
• Agent automatically verifies installation by sending a "Hello" message.

Open Claw Agent Configuration and Verification

• Open Claw installed and Gemini installed.
• Agent prompts to connect a messaging channel.
• Agent verifies its own setup by opening a terminal and sending a "Hi" message.
• Agent opens the Open Claw dashboard as requested.
• Agent updates its identity and persona (e.g., "You are my personal assistant. I will call you Jerry. You are our funny assistant. Always speak politely.").

Messaging Channel Setup (Discord)

• Agent is tasked with setting up a messaging channel, prioritizing Discord.
• Agent plans to integrate Discord, requiring user approval for steps.
• User provides the Discord bot token.
• Agent configures the Discord channel and DM policy.
• User is prompted to send a message from Discord to confirm pairing.
• Agent receives a pairing code from Discord and the user provides it back to the agent.
• Agent successfully pairs with Discord.
• Troubleshooting is required when Discord messages are not immediately processed.
• Agent analyzes logs to diagnose issues.
• The issue is resolved, and Discord messages are processed.

Agent vs. Chatbot Functionality

• A key distinction is made between chatbots (answering questions) and agents (completing tasks/goals).
• Agents can access and utilize tools and skills to achieve goals.
• The demonstration focuses on proving the agent's capability beyond simple chat.
• Task: Send a message from the phone to the agent to set up live gateway logs.
• Agent's goal is to achieve the task, using available tools or skills.

Tool and Skill Utilization

• Initially, the agent uses a built-in "weather" skill to answer a query about Karachi's temperature.
• When the "weather" skill is disabled, the agent defaults to using the "web search" tool.
• When both skills and web tools are disabled, the agent utilizes a `curl` command to retrieve information.
• This demonstrates the agent's proactivity in achieving its goal using various methods.

Setup and Initialization

• The video demonstrates setting up and running Open Claw with a General Agent for AI workflows.
• Key components include a Gateway that handles incoming traffic and directs it to the appropriate agent.
• The process has shifted from manual setup to a more automated, three-step approach involving a user, a General Agent, and Open Claw.
• A provided zip file contains an `agent.md` file which acts as the "brain" for the General Agent, outlining tasks related to Open Claw.

General Agent Workflow

• The General Agent reads the `agent.md` file to understand its capabilities and tasks.
• When asked "What can you do for Open Claw?", the agent references its instructions to list capabilities like installation, configuration, and automation.
• The user initiates tasks by asking the General Agent to create a plan, which the agent then generates based on its instructions and documentation.
• The user reviews and approves the plan, after which the agent executes the steps.

Automated Open Claw Setup

• The General Agent can automate the installation and configuration of Open Claw.
• It checks for existing installations, sets up API keys (like Gemini's), and performs onboarding steps.
• The agent requires user approval and potentially API keys to proceed with the setup.
• The process now involves the agent handling tasks that were previously done manually, such as running commands and configuring channels.
• A crucial step for channel connection (e.g., Discord) requires the user to create the bot on the platform and provide its ID to Open Claw, after which the agent can connect and reconfigure it.

Daily Tech News Agent

• Automates daily tech news gathering from specified sources.
• Saves time by consolidating information into a structured report.
• Agent schedules and executes the task daily at a precise time.
• Provides concise summaries with the significance of each development.

System Health & Readiness Report

• Automates operating system checks (OS, CPU, memory, disk space).
• Verifies availability of specific software like Docker and Ollama for AI workloads.
• Agent interacts directly with system tools for data collection.
• AI reasoning verifies AI-generated output against native OS commands.

Specialized Technical Research Workspace

• Creates dedicated workspaces for specialized agents (e.g., technical research).
• Configures bootstrap files, specialized agents, and user preferences per workspace.
• Agent role and goals are defined in markdown files for consistent behavior.
• Generates comprehensive reports in markdown and PDF formats.
• OpenClaw automatically handles missing package installations (e.g., pandoc).

Marketing Operations Monitoring

• Monitors official marketing channels (website, social media, video).
• Reviews past week's marketing activity and generates structured reports.
• Utilizes web interface for configuration and execution.
• Workspace configuration allows for simple, task-focused prompts.
• Reports include an executive summary and are organized by content type.

Open Source Agent Framework

• OpenClaw revolves around agents, which are powered by AI models, can use tools and skills, and communicate through channels.
• Interactions are stored as sessions.
• OpenClaw's capabilities are extended through tools, skills, plugins, channels, and automation hooks.

Channels and Gateway

• Channels are the communication layer between users and OpenClaw agents (e.g., Telegram, Slack, Discord).
• Channels only forward messages to the OpenClaw gateway; they don't process AI requests.
• The gateway is central, receiving messages, authenticating users, routing requests, managing sessions, and returning responses.

OpenClaw Web Interface vs. Channels

• The OpenClaw web interface is for administrators and developers for configuration, testing, and debugging.
• Channels allow agents to be accessible on platforms users already use (e.g., Telegram, Slack).

Agent Workspace and Configuration

• Each agent has a workspace directory for tools and context.
• Key files in the workspace include:
  • agents.md: Agent's operating instructions and guidance.
  • soulm.md: Agent's personality and communication style.
  • identity.md: Agent's identity and name.
  • user.md: Information about the user for personalized responses.
  • tools.md: Notes and conventions for using tools.
  • memory.md (optional): Agent's long-term memory.
  • bootstrap.md: Used for initial workspace setup.
• openclaw.json: Contains overall OpenClaw environment configurations.

Tools, Skills, and Hooks

• Tools allow agents to perform actions beyond text generation (e.g., search, execute commands).
• Skills are markdown files teaching agents how and when to use tools, acting as reusable workflows.
• Hooks enable automatic responses to events (e.g., notifications after a deployment).

OpenClaw CLI Commands

• openclaw doctor: Performs health checks and suggests fixes for components.
• openclaw gateway status: Shows the status of the gateway service.
• openclaw logs -follow: Tails gateway log files.
• openclaw --version: Displays the OpenClaw version.

Configuring Telegram Channel

• Create a Telegram bot using BotFather and obtain a bot token.
• Use the openclaw channels add command to configure the Telegram channel.
• Paste the bot token when prompted.
• Select the agent to handle incoming messages (usually the default agent).
• In Telegram, start the created bot.
• Copy the provided command and paste it into the OpenClaw SSH terminal for pairing.
• Messages sent to the Telegram bot are processed by OpenClaw and responses are sent back.

Introduction to Agents and Open Call

• Agents are systems designed to act on behalf of a user.
• Open Call and Hermes are platforms for building AI agents on NVIDIA Jetson.
• These agents can access tools, interact with local files, and perform automated tasks.

Key Concepts: Memory and Skills

• Memory: Agents can store and retrieve past interactions for long-term context (e.g., memory.md file with embeddings for semantic search).
• Skills: Repeatable workflows that teach agents how to perform specific tasks deterministically.
• Jetson Agent Skills can automate Jetson BSP setup or optimize applications directly on the device.

Nemo Crawl: The NVIDIA Agent Stack

• Nemo Crawl is NVIDIA's reference stack for agentic harnesses like Open Call and Hermes.
• It integrates NVIDIA Open Shell for secure sandboxing, restricting agent access and actions.
• Jetson OS (e.g., Jetpack 7.2) comes with Nemo Crawl support for easy installation and local/cloud model execution.

Live Demonstrations and Security

• Demos showcased agents performing interactive quizzes, controlling robot arms, and interacting via Telegram.
• NVIDIA Open Shell acts as a security sandbox, preventing unauthorized file system access or actions.
• Policies define allowed agent actions, ensuring a zero-trust environment.

Hermes vs. Open Call and Use Cases

• Hermes: Excels at specialized tasks like research and coding assistance, with self-improving capabilities.
• Open Call: Offers extensive tool integration (e.g., WhatsApp, Discord) and a wide range of skills.
• Both can be used for industrial AI automation, summarizing assembly lines, and gathering exception information.

Tips for Jetson Users

• For limited memory (e.g., Jetson Nano 8GB), use optimized models (like Gemma 2B) and inference frameworks (like Llama CPP).
• Consider using cloud models or API calls if local resources are insufficient.
• Define clear use cases to leverage Jetson's strengths, like local data processing and security.

Future and Community

• Upcoming episodes will delve deeper into connecting robots with agents.
• Resources for learning more include Discord, GitHub, and the Jetson AI Lab website.
• NVIDIA will be present at GTC for further engagement.

Understanding Claw Machines

• Claw machines are designed so the claw intentionally weakens, giving a low chance of winning.
• The machine's win rate is configured by the owner, not determined by skill.
• In Japanese arcades, claw machines generate 59% of revenue, significantly more than video games.

The Business Model: Costs and Revenue

• Machine Cost: New machines range from $700-$10,000; secondhand ones are cheaper.
• Prizes: Bulk plush toys cost about $1 each, creating a profit margin on the perceived value.
• Location Cut: Venues take 10-50% of earnings for machine placement.
• Restocking & Collection: 10-25% of takings cover maintenance and coin collection.
• Profit: A well-placed machine can earn $300-$700+ per month, paying for itself in 3-6 months.
• Margins: Typically run 50-70%.

Legal Considerations and Risks

• Machines must comply with state regulations, often requiring non-monetary prizes below a value cap.
• Exceeding chance thresholds can lead to machine seizure and fines.
• Operators must understand and adhere to rules to avoid legal trouble.

The Psychology of the Sale

• Customers are not buying the toy, but the "hope" or the 20 seconds of perceived winning opportunity.
• The "wanting" or desire to win is the actual product being sold.

Introduction to AI Agent Memory Issues

• Current AI agents forget between turns, relying on pasting entire histories into prompts, leading to massive token usage.
• Larger context windows don't solve the problem; accuracy degrades before they are full.

Tencent's Open-Source Memory Solution

• Tencent released an MIT-licensed, locally runnable memory system aiming to improve agent performance while reducing token use.
• It claims a 50% pass rate increase and over 50% token reduction on a benchmark.

Core Concepts: Episodic vs. Semantic Memory

• Inspired by 1972 psychology, the system differentiates between remembering events (episodic) and distilled facts (semantic).
• Forgetting is reframed as a necessary compression step, not a bug.

Idea 1: Symbolic Short-Term Memory (Task-Specific)

• Full tool outputs are moved out of the prompt into markdown files.
• A mermaid diagram (boxes and arrows) summarizes the workflow, with nodes linking back to raw data via IDs.
• Compression is threshold-driven (50% mild, 85% aggressive) and the diagram has a token budget.

Idea 2: Layered Long-Term Memory (User-Specific)

• A four-layer pyramid: L0 (raw conversation), L1 (atoms/facts), L2 (scenes/tasks), L3 (persona/habits).
• The persona is rebuilt periodically (default: every 50 new memories).
• Recall prioritizes layers from top (persona) down to raw conversation.

Benchmark Analysis

• The headline benchmark shows a pass rate increase from 33% to 50% with token use dropping from 221M to <86M.
• One row (software engineering) slightly overstates token reduction (printed 33.09% vs. calculated 31.6%).
• Performance gains are larger on bookkeeping tasks than complex reasoning.

Technical Implementation and Cost

• Defaults to SQLite with a vector extension running locally.
• Retrieval is hybrid keyword and vector search.
• A 5-second timeout on recall prevents stalling.
• It requires patching the agent runtime and has a maintenance obligation after host upgrades.
• The project has an open-core model: local default with a managed cloud backend option (Tencent's DeepSeek model).

Version 2: Shared Memory Assets

• Memory evolves from a private cache to shared assets: Chat memory, Skills, Wiki, Code Graph.
• Skills are versioned, resource-aware procedures, generated from successful agent runs.
• A wiki allows agents to write and maintain linked markdown entity pages.
• A code graph indexes codebase connections, allowing pre-population for new agents.
• These assets have owners, versions, and permissions (private by default).

Unaddressed Problems

• Caching Bill: Shifting prompt prefixes due to memory injection can reduce API prompt caching benefits, increasing costs.
• Attack Surface: Malicious content can be consolidated into long-term memory, requiring periodic audits.

Recommendations

• For supported runtimes: Yes, especially for long sessions. Start with long-term memory, add compression later. Monitor provider bills for caching effects.
• For others: Steal the architecture (file-based context, IDs, periodic distillation).

Future Roadmap: Portability

• The crucial unchecked box is "Portable memory" (cross-agent, cross-framework, cross-device).
• Without portability, agent memory becomes locked into a framework, creating vendor lock-in.

Introduction

• This video demonstrates a speedrun of a standing doll capture in Tokyo Claw Machine on Roblox.

Technique and Tips

• The primary strategy involves precisely positioning the claw.
• The goal is to hook the doll's head or upper body for a secure grip.
• Patience and careful observation of the doll's resting position are crucial.
• The video highlights the importance of understanding the claw's physics within the game.

Outcome

• The speedrun successfully captures the standing doll.
• The focus is on efficiency and quick execution.

Initial Fishing Attempts

• Started by using a crane to catch a fish (Nemo).
• Upgraded from a basic log boat to a fishing boat and a steel fishing claw.
• Experienced dangerous waves and tsunamis.

Upgrading and Catching Bigger Fish

• Purchased a better boat (fishing boat) and a stronger fishing crane.
• Successfully caught a dolphin (90 kg) and a sunfish (400 kg).
• Upgraded the aquarium to accommodate larger fish.
• Acquired a yacht and later a sailboat.

Advanced Catches and Maximum Earnings

• Focused on catching mutated or silver fish for higher value.
• Caught a silver dolphin and a golden sunfish.
• Acquired a container ship and then the "yak claw" for deeper fishing.
• Caught legendary white sharks, a treasure shark, and a mythical great white shark.
• Successfully caught a Megalodon and a whale shark.
• Reached a high income of 100k per second.
• Maxed out the aquarium size.
• Stole fish from a neighbor's aquarium.

Project Overview

• OpenClaw is an open-source, self-hosted AI agent framework that runs locally on devices (macOS, iOS, Android, Windows via WSL 2).
• It flips the traditional cloud-first AI model by keeping sensitive data on local hardware.
• Functions as a customizable robot butler, executing digital chores in the background, directly interacting with apps like WhatsApp and Slack, and controlling web browsers for tasks like form filling.

Rapid Growth and Acquisition

• Launched in late 2025, the project saw unprecedented growth, reaching over 190,000 GitHub stars in its first 14 days.
• It surpassed major projects like the Linux kernel and React in GitHub stars.
• Creator Peter Steinberger was hired by OpenAI to build their next generation of intelligent agents.
• After Steinberger's departure, the project's momentum accelerated, continuing to gain stars and solidifying its position.

Significance and Future of AI Agents

• OpenClaw solved the bottleneck of AI agents requiring direct local system access to execute tasks.
• It enables AI to bridge the gap between text generation and physical file manipulation by granting system-level action capabilities.
• The MIT license allows for free use and development, fostering a decentralized, user-controlled ecosystem.
• The project signifies a shift from prompt engineering to autonomous delegation and highlights the industry's focus on AI that can drive computers, not just generate text.
• Running locally offers privacy and control but comes with a hardware performance trade-off.

Introduction to OpenClaw Platinum

• The presentation introduces OpenClaw Platinum, an AI-powered business automation solution.
• The speaker, Chad Nicely, highlights its potential to plan, build, and sell businesses using voice commands.
• He emphasizes that this technology makes success accessible regardless of prior experience.

Core Functionality and Demonstration

• OpenClaw utilizes AI models like Claude (Anthropic) to automate business tasks.
• A live demonstration shows OpenClaw generating business ideas, planning product architecture, writing sales copy, and deploying sales pages.
• The AI can create video sales letters, integrate with payment gateways (Stripe), and deploy traffic assets.
• The system progresses through seven stages of business creation, from idea generation to scaling plans.

Key Features and Benefits

• Agentic AI: OpenClaw moves beyond simple answering to executing complex goals and proactively fixing issues.
• Digital Workforce: It acts as a team of AI workers managed by a CEO (the user), handling tasks like customer support, campaign drafting, and lead generation.
• Cost-Effectiveness: The solution operates at a fraction of the cost of hiring human VAs, handling tasks for pennies on the dollar.
• Automation of Tasks: It automates repetitive tasks such as checking revenue, updating pages, sending follow-ups, and managing ads.
• Security: OpenClaw operates in a controlled environment to protect the user's core business, avoiding direct access to personal accounts or financial information.

The OpenClaw Platinum Offer

• Platinum Package: Includes everything in Silver and Gold, plus Chad Nicely's entire software suite (Mint Bird, Page Sprout, Course Sprout, Letterman, Quiz Forma, Global Control).
• Software Suite: Lifetime access to tools for creating sales pages, funnels, courses, newsletters, quizzes, and managing contacts/workflows.
• Agency License: Platinum members receive 200 installs of Claw Launcher, allowing them to deploy OpenClaw for clients.
• Bonuses: For purchasing on the call, bonuses include Titan Layer subscription, Postream subscription, and an affiliate link for the next launch.
• Training and Support: The offer includes a 7-day intensive workshop, access to a members area, skills vault, command center, and dedicated support channels (WhatsApp, Telegram).

Business Opportunity and Future

• The AI market is seen as a "gold mine" opportunity for the next 12-18 months due to rapid advancements and user adoption.
• Users are encouraged to offer free installs to clients, then upsell onboarding services or ongoing management.
• The speaker predicts that software will become free in the future, with AI and skills being the primary revenue drivers.
• The focus on local newsletters is highlighted as a high-potential asset with millions in acquisition value.

Game Mechanics

• Start by entering a boat and driving it to the ocean.
• Use the crane to catch fish by clicking when the crane is near them.
• Fish caught can be placed in an aquarium to generate money.
• Money can be used to upgrade the boat, aquarium, crane, and claw.
• Waves can damage or kill the player.
• Different fish rarities (common, rare, epic, legendary, mythical, rainbow) exist, with higher rarities yielding more money.

Upgrades and Progression

• Early upgrades include a better boat and crane.
• Purchasing a "fishing claw" is recommended over a "cruise claw".
• Aquarium upgrades increase fish capacity and income.
• Boat upgrades increase speed, which is crucial for avoiding waves and catching faster fish.
• Crane upgrades increase the reach or depth.
• Claw upgrades improve the ability to catch heavier or rarer fish.
• Players can AFK to passively earn money.

Notable Catches & Goals

• Caught: Clownfish, Koi, Gold fish, Silver Bass, Dolphin, Rare Tuna, Rainbow Dolphin, Sunfish, Gold Pufferfish, Gold Bass, Rainbow Pufferfish, Moon Jellyfish, Stingray, Rainbow Shark, Legendary Whale, Mythical Tiger Shark, Rainbow Jellyfish, Golden Epic, Silver Swordfish, Rainbow Sunfish, Golden Swordfish, Whale Shark, Rainbow Stingray, Sea Turtle, Gold Shark.
• Goal: Catch mythical fish.
• Rainbow fish are noted to be as valuable as legendaries.
• The player aims to save up for increasingly expensive boats and claws to reach further into the ocean.

End Game & Frustration

• The player struggles to catch mythical fish despite reaching high levels of upgrades.
• A "mythical tiger shark" got away, causing frustration.
• The player expresses a desire to catch a mythical fish in a future attempt.
• The final boat, a yacht, costs 30 million.

Introducing Hyperagent

• Hyperagent is a new AI agent platform where each agent gets its own cloud computer, enabling them to perform tasks in the real world rather than just describe them.
• It was developed by the team at Airtable as an internal tool before becoming a product.
• The process of creating an agent is simplified by describing the job in a few sentences, with the platform asking clarifying questions to fill in the gaps.

Building the "Scout" Agent

• The author demonstrates building a competitive intelligence analyst agent named "Scout."
• Scout's setup involves defining its identity (system prompt), selecting a model (e.g., GPT), and enabling tools like web search and document saving.
• Each agent includes a built-in browser for direct interaction with websites.

Agent Capabilities: Skills and Memory

• Skills: Agents can be taught reusable skills, such as how the user personally judges competitor moves, allowing them to reason and act consistently. These skills can be shared across agents.
• Memory: Agents can auto-save important information, preventing them from starting from scratch each session and improving their performance over time by remembering past findings and user preferences.

Live Mode and Integration

• Live Mode: Agents can be set to run autonomously and report back via Slack.
• This feature allows agents to monitor specified companies or markets and post updates only when something significant changes, operating in a "silent unless it matters" mode.
• Scout successfully identified and analyzed a major market move by Google, providing actionable insights for the author's company.

Team of Agents

• The platform supports building a team of agents that can collaborate and pass work between themselves.
• Examples include an inbox agent for prioritizing tasks and a content agent for generating posts.
• Agents are designed to handle routine tasks, escalating judgment-based decisions to humans for approval.

Command Center and Real-World Applications

• A command center allows users to monitor agent activity and score their performance to continuously improve them.
• Airtable: An agent answers questions across a 70,000+ entry data catalog in Slack, saving 200 hours weekly.
• Deskimo: An eight-agent sales pipeline was built in a week, automating prospecting and outreach.
• Deskimo's agents learn from founder feedback, improving outreach email effectiveness significantly.

Hyperagent vs. Hermes/Openclaw

• Hermes is open-source, requiring self-hosting and maintenance, appealing to those who want full ownership and privacy.
• Hyperagent offers a managed service where each agent gets its own cloud computer, focusing on getting work done without the user needing to manage infrastructure.
• For the author's needs, Hyperagent proved to be the stronger solution.

Agent Runtimes Explained

• Personal agent runtimes are like background services (demons) that use language models to reason.
• They run on your server, can access your files, remember past interactions, and operate continuously.
• This is unlike chatbots like ChatGPT or Claude which lose context when closed.

OpenClaw's Rise and Fall

• OpenClaw gained immense popularity quickly, reaching 380,000 GitHub stars.
• However, it suffered from severe security vulnerabilities, including remote code execution and numerous CVEs.
• Its marketplace, Claw Hub, was found to host thousands of malicious skills.
• Updates frequently broke existing setups, leading to user dissatisfaction.

Hermes Agent's Approach

• Hermes Agent, developed by Nanos research, focuses on self-improvement.
• It learns from completed tasks by writing and loading markdown skill files.
• A migration command is available to transfer OpenClaw setups to Hermes.

Criticisms of Hermes Agent

• Users report Hermes can be overly optimistic about its performance, potentially enshrining errors as skills.
• Memory management can be inconsistent, with some users preferring OpenClaw's more transparent markdown files.
• Concerns were raised about a default web search feature sending data to a company where the developer worked, with a swift merge by the founder and lack of reviews.
• There are suspicions of astroturfing and some developers refuse to use it on principle due to moderation issues.

Key Differentiators and Future

• Both platforms have security flaws, with the OS being the main defense.
• Hermes users showcase impressive self-built agent capabilities, like daily summaries and inbox triaging.
• Major platforms (OpenAI, Anthropic, Google) are integrating similar agent features, potentially making independent runtimes obsolete.
• Open-source and model-agnostic nature are key advantages of indie runtimes.
• Hermes' business model involves selling tokens, with significant investor interest.
• Reliability and the ability to survive updates are highlighted as key drivers for users switching from OpenClaw to Hermes.

AI Business Creation with OpenClaw

• AI can generate a complete business from idea to execution in minutes.
• The process starts with simple commands like "{forward-slash}create business".
• AI identifies potential business models, recommends one (e.g., Etsy digital products), and builds a blueprint.
• It generates sales copy, designs a sales page, creates a video (including voiceover), and sets up Stripe integration.
• The AI also creates traffic and conversion assets like organic posts, ad hooks, and email sequences.
• It can deploy content to social media platforms for organic reach and offers options for paid ads and email campaigns.
• The final stage involves optimization and scaling plans, including A/B testing headlines and adding testimonials.

OpenClaw Technology

• OpenClaw is the AI technology used for this automated business creation.
• It requires a contained environment for safe operation.
• Proper setup, onboarding, and usage are crucial for success.
• A "Claw Launcher install," "Fast Start Members Area," and onboarding process are key components.
• The developers claim to have "cracked the code" for using OpenClaw effectively.

Impact and Future

• AI is democratizing business creation, making it accessible to anyone regardless of experience.
• The technology is rapidly evolving, with AI becoming more capable over time.
• The next 12-18 months are presented as a pivotal period for online careers due to AI advancements.
• The AI itself won't take jobs; people who can manage and utilize AI will be the ones succeeding.

OpenClaw Setup

• Download all commands and scripts from the provided website.
• Install OpenClaw on Windows using PowerShell: docs.openclaw.ai/install.
• Use "Quick start" setup mode and authenticate with an AI provider (e.g., OpenAI, Claude).
• Skip unnecessary channels and skills initially, but enable "hooks" like session memory.
• The gateway is the core running process for OpenClaw.
• Verify OpenClaw version: requires 2026.6.9 or newer; use openclaw --update if needed.
• Test OpenClaw by sending a "Hi" message in the dashboard.

OcuClaw Skill Installation

• Name your agent (e.g., Jarvis).
• Instruct OpenClaw to install the OcuClaw skill. Be patient, as this can take several minutes and may show errors or appear stuck. Create a new session if a "session conflict" error occurs.
• Create a relay token for secure communication.
• Ensure PowerShell execution policy allows local scripts.

Secure Connection with Tailscale

• Install and sign in to Tailscale to create a private network between devices.
• Verify computer connection in Tailscale.
• Inform OpenClaw that Tailscale installation is complete.
• OcuClaw is an officially sanctioned skill of OpenClaw.

Connecting Phone and G2 Glasses

• Install OcuClaw via the Even Reality app (Evenhub).
• In OcuClaw settings, configure the relay server address (WSS + port 8444) using your Tailscale hostname.
• Enter the previously created relay token.
• A successful connection shows OcuClaw connected with OpenClaw session and model info.

Enabling Debugging

• Recommended for generating bug reports.
• Run specific PowerShell commands to enable opt-in debugging.
• Restart the OpenClaw gateway.
• In the OcuClaw app, go to settings -> display -> debug and toggle it on.
• Use the bug icon to report issues with context.

Voice Integration with Sonics

• Test the core connection by sending a message from the OcuClaw app and seeing the response in the G2 glasses.
• Sign up for Sonics (sonix.com) for speech-to-text, add payment, and create an API key.
• Configure OcuClaw voice settings to use Sonics with your API key.
• Restart the OpenClaw gateway after updating Sonics API key.

Final Test and Capabilities

• Activate listening in OcuClaw via glasses controls (double-tap ring/touchpad, then select listen).
• Speak a prompt; the response should appear in the glasses.
• Manual listening activation/deactivation is useful for privacy.
• The setup allows the AI agent to interact with local computer tools, documents, and services.

Claw Fishing Gameplay

• The player is playing a game called "Claw Fishing" and aims to catch rare sharks like the Rainbow Great White.
• Successfully catches a Whale and a Rainbow Hammerhead Shark.
• Obtains a Rainbow Tiger Shark and then a highly valuable 400,000 point catch.
• Discusses the cost of private servers (Robux) and player count for the game.

Rare Catches and Upgrades

• Catches a Rainbow Tiger Shark and aims to catch another.
• Discusses the difficulty and reward of catching a Megalodon.
• The player gets a Megalodon, noting its size compared to a whale and a great white shark.
• They encounter and attempt to catch a Kraken.
• The player mentions buying the best crane in the game.

Other Games

• The players briefly switch to a game called "Slap to Survive" where they fight zombies.
• They discuss Pokemon cards and a "stack" of cards.
• In "Slap to Survive", they use a "home run" mechanic to defeat zombies.

Podcast Introduction

• The OpenClaw Podcast, episode 6, features a Q&A session with hosts Hannis Rudolph and Patrick.
• A guest from Microsoft was expected but rescheduled for the following week.

Favorite OpenClaw Use Cases

• Personal Use: Automating daily life tasks, like planning hikes and dinners, through WhatsApp integration to help fiancées become more "agentic" in a casual way.
• Business Use: Automating customer text messages and emails for a side business, updating notes automatically to save time.
• Family Business: Running marketing for a parent's small boat cover business and implementing a five-step escalation strategy for a customer service issue.

Upcoming Releases and Stability

• The previous release focused on UX and onboarding but caused initial issues for some users.
• The team is focused on reducing unnecessary configurations and reworking the CI process to prevent breaking updates.
• OpenClaw is about 9 months old, and the current focus is on stability and "paying down early taxes."
• Significant improvements are expected in the next release, described as "OpenClaw version 2.0."

Mobile Apps and Future Features

• Improvements to the mobile apps are tied to the upcoming release, which includes backend rework for better usability.
• The decision to release mobile apps early was to gather feedback and attract new contributors, leading to rapid iteration.
• Real-time voice interaction is a planned feature, with ongoing work on FaceTime support.

ClawHub and Security

• ClawHub is the marketplace for OpenClaw skills and plugins.
• Claw Scan, an open-source scanner, was developed based on learnings from scanning thousands of skills monthly to address security concerns.
• ClawSpecs, sharable agent specifications, are planned for release to provide reusable templates for building agents.
• OpenClaw's ClawHub actively rejects skills containing API keys to prevent security breaches.

Open Weight Models and Security Concerns

• The discussion touches on the AI leadership letter and the debate around open-weight models versus proprietary ones regarding safety.
• There's a narrative that corporate models are naturally safer and more resistant to prompt injection than open-weight models.
• The practical feasibility of using open-weight models for personal agent tasks is increasing due to advancements in model capabilities.
• The high token costs of frontier models are a factor, making open-weight alternatives attractive for personal use.
• Users are advised against sending passwords or account credentials directly to AI; secure methods like requesting credentials via password managers are recommended.

Technical Updates and GUI

• Scheduled tasks, script actions, and crons are being unified, potentially removing the need for a separate heartbeat mechanism.
• A GUI application for OpenClaw is being reworked as part of the onboarding and UX overhaul.
• The goal is to simplify the download and installation process, moving towards a user-friendly website experience.

Challenge 1: Image Upload Service

• Hermes Agent: Discovered the vulnerable library (IM Magic) and achieved remote code execution (RCE) by exploiting a CVE. Awarded 10/10.
• Agent Zero: Identified some services but failed to find the vulnerable library or exploit it. Awarded 3/10.
• Open Claw: Discovered IM Magic and SSRF, but not full RCE. Awarded 5/10.

Challenge 2: Port Scanning & Service Identification

• Agent Zero: Successfully identified all open ports and services, found a CVE, and validated it, though it refused to execute a command. Awarded 8/10.
• Hermes Agent: Discovered limited ports and a denial-of-service vulnerability, but not the full exploit path. Awarded 3/10.
• Open Claw: Failed to discover all ports and services, performing poorly. Awarded 1/10.

Challenge 3: Comfy UI Vulnerability

• Open Claw: Achieved RCE quickly. Awarded 10/10.
• Hermes Agent: Achieved RCE by modifying configurations. Awarded 10/10.
• Agent Zero: Confirmed vulnerability but did not validate with RCE. Awarded 7/10.

Challenge 4: N8N Multi-Exploit Chain

• Hermes Agent: Achieved RCE through an alternative path, not chaining exploits as intended. Awarded 10/10.
• Agent Zero: Was blocked and unable to exploit the target. Awarded 1/10.
• Open Claw: Was blocked and unable to proceed.

Overall Winner

• Hermes Agent: The overall winner, scoring 10/10 on most challenges.
• Agent Zero excelled in port scanning and service identification but struggled elsewhere.
• Agentic frameworks' performance varies, with no single "best" for all tasks.

Gameplay Overview

• The video showcases gameplay from the Roblox experience "Claw Fishing."
• The player attempts to operate a claw machine to grab prizes.
• The gameplay appears challenging, with moments of near success and failure.

Player Experience

• The player expresses anticipation and frustration throughout the gameplay.
• There are audible exclamations of surprise and dismay.
• The overall tone suggests a difficult but engaging experience.

Introduction to OpenClaw

• OpenClaw (or Open Cloud) is an AI assistant that acts as a personal assistant, fully managed by AI.
• It can handle various tasks such as managing finances, scheduling meetings, conducting research, and even acting as a virtual partner.
• The primary goal is to assist users, making tasks easier and more efficient.

How OpenClaw Works

• OpenClaw is a software that needs to be installed on a device (Mac, Linux, Windows).
• It requires an AI model (e.g., Google Gemini, OpenAI Codex, Claude) to function.
• Interaction is seamless through platforms like WhatsApp and Telegram, making it easily accessible.
• Upon installation, the AI agent is "blank" and needs to be trained and given a role, similar to hiring a new employee.

Capabilities and Use Cases

• OpenClaw can be trained for various roles: personal finance management, coding assistance, sales support, and administrative tasks.
• It has advanced capabilities due to its access to numerous tools and skills that can be installed.
• Users can install specific "tools" or "skills" to enhance the agent's abilities (e.g., presentation creation, financial analysis, prospecting).
• Examples include "Devi" used for event management and personal assistance, "Ay" for personal assistance, and "Usman" for managing mosque affairs.
• The AI can "self-clone," creating new agents with specific skills based on existing ones.

Security and Mitigation

• Concerns about security arise if installed on devices with sensitive data.
• Analogy: Hiring a person with access to confidential information.
• Mitigation strategies include:
• Installing on devices with limited sensitive data (like a Raspberry Pi).
• Not providing the AI with any confidential information.
• Being cautious as agents can be "tricked" or "injected" with malicious scripts.

Advanced Features and Benefits

• OpenClaw agents diligently record interactions, enabling them to understand context over time, similar to a human assistant taking notes.
• This note-taking allows for shorter, more effective prompts as the AI remembers past interactions and preferences.
• The AI can connect to business systems like ERPs (e.g., MCP, Odoo) to automate reporting (e.g., daily revenue) and identify hot sales pipelines.
• It can even handle operational tasks like polling 40 meeting attendees about their food preferences within budget.
• AI agents can seamlessly blend into business processes, with clients often unaware they are interacting with an AI.

Tips for Implementation

• Focus on a specific scope for each AI agent; a broader scope can lead to less effective performance.
• Understand the need for AI models and be prepared for potential costs associated with premium models.
• Clearly define the use case for the AI to maximize benefits and avoid wasted resources.

Game Mechanics

• Players drive a boat out to sea to fish using a crane and claw.
• Fish are caught by lowering the claw and clicking to grab them.
• Caught fish can be placed in an aquarium, which generates in-game currency over time.
• Upgrades include the boat, aquarium (for more cash or space), and crane/claw (to catch heavier fish and go deeper).
• Tsunamis can appear and kill players if they are caught in them.

Progression and Upgrades

• Early game involves catching common fish like clownfish and bass to earn money.
• Upgrading the boat allows players to travel further out to sea.
• Upgrading the claw and crane is crucial for catching rarer and heavier fish.
• VIP passes can be purchased for Robux, though their benefits are not explicitly stated.
• Mutated fish significantly increase income per second compared to regular fish.

Rare Catches and Goals

• The goal is to catch increasingly rare and valuable fish, such as mythical tiger sharks, rainbow dolphins, and legendary whale sharks.
• Mutations are key to high income; regular legendary fish provide less income than mutated epics.
• Players progress towards affording the most expensive upgrades, like the "Yacht" tier boats, cranes, and claws, to reach the deepest parts of the ocean.
• The ultimate goal appears to be reaching the furthest possible fishing spots and catching the rarest mythical or rainbow variants.

Player Experience

• The player expresses frustration with missing rare fish and the difficulty of controlling the claw precisely.
• Controlling the claw with arrow keys after it's deployed is discovered mid-game, making fishing easier.
• The player notes that mutations are more valuable than rarity for income generation.
• The game requires significant grinding or Robux investment for high-tier upgrades.

AI Operating System Concepts

• AI is evolving from isolated tools to integrated systems that continuously think, act, and improve.
• An AI Operating System (AIOS) combines agents, memory, tools, and interfaces for end-to-end workflows.
• Evolution phases: Tools (isolated), Workflows (sequential automation), Agents (decision-making), AIOS (integrated systems).
• Key differences from traditional systems: Intelligence (reasoning vs. automation), Adaptability, Interconnectedness, Persistence (always on vs. triggered).

Core Components of an AI Operating System

• Agents: Reasoning engines that plan, prioritize, and execute decisions.
• Memory: Retains context across interactions (short-term, long-term, vector) for learning and personalization.
• Tools: Execution layer that allows interaction with the external world (APIs, code execution).
• Interfaces: Connects the system to users and other systems (chat, API, voice).

Designing Intelligent Agents

• Agent behavior is designed, not emergent. Reliability comes from architecture.
• Key mechanisms: System prompts (identity, goals, constraints, tone), Task Decomposition (breaking down complex tasks), Structured Reasoning (Plan-Act-Observe-Iterate loop).
• Role design: Specialized agents (planner, executive, reviewer) improve performance and modularity.
• Multi-step thinking is essential for complex problems, using iterative loops rather than single steps.

Multi-Agent Systems and Orchestration

• Multi-agent systems distribute responsibilities for collaboration, specialization, and complex goal achievement.
• Advantages: Divide and conquer, parallel execution, specialization lead to faster, more accurate, and scalable systems.
• Patterns: Sequential, parallel, and hierarchical collaboration.
• Orchestration manages agent coordination, task routing, communication, and flow control.
• Supervisor-worker architecture provides centralized control for simpler coordination and better observability.

Memory and Retrieval Systems

• Memory enables context-aware behavior, personalization, multi-step reasoning, and continuity.
• Layers: Short-term (context window), Long-term (persistent knowledge), Persistent Storage (databases, vector stores).
• Retrieval is crucial for activating memory, using semantic search (embeddings, vector databases) to find relevant information.
• Embeddings are numerical representations capturing semantic meaning for efficient search.

Execution Layer and Workflow Automation

• Tools bridge thinking and doing, enabling AI systems to perform actions via APIs, code, etc.
• Structured inputs/outputs and clear function calling patterns are essential for reliable execution.
• Workflow automation connects tools into structured pipelines for end-to-end execution.
• Orchestration manages workflow execution, dependency management, and error handling.
• Key strategies: Retry logic, fallback mechanisms, logging, and reliability by design.

Interfaces and Integrations

• Interfaces (chat, API, UI, voice) connect users and systems to AI, hiding complexity and maximizing usability.
• APIs provide programmatic access for integration into applications and systems.
• Web hooks enable event-driven automation, reacting instantly to external events.
• Integrations with databases, APIs, and other services are crucial for real-world impact and autonomous systems.

Continuous Execution and Autonomous Agents

• Continuous AI systems operate indefinitely, looping, observing, and acting without constant human input.
• Event-driven agents are triggered by system events, enabling real-time responsiveness.
• Persistent workflows maintain state across runs, allowing for long-term execution and recovery from interruptions.
• Autonomous agents operate through a continuous loop of observation, reasoning, and action, enabling self-improvement.

Feedback Loops and Optimization

• Feedback loops enable continuous learning, improved accuracy, error reduction, and system optimization.
• Types of feedback: Explicit, implicit, system, and combined.
• Iterative improvement involves prompt refinement, parameter tuning, and workflow updates.
• Adaptive systems adjust in real time based on personalization, context, and feedback.
• Monitoring and measurement are critical for understanding performance and driving optimization efforts.

Debugging and Reliability

• AI failures are often subtle (hallucinations, reasoning errors, context issues) and hard to reproduce.
• Key debugging strategies: Logging (capturing all inputs/outputs/decisions), Observability (monitoring metrics, dashboards, alerts), Testing (unit, integration, scenario-based), Workflow Inspection & Isolation.
• Failures are opportunities for improvement, leading to more robust and reliable systems.

Optimization and Production AI

• Optimization balances speed, accuracy, cost, and scale through trade-offs.
• Key drivers: Token usage, model selection, tool/API calls, and infrastructure.
• Strategies: Token efficiency (prompt reduction, context management), Latency reduction (parallel execution, streaming), Caching, and choosing appropriate models.
• Continuous monitoring is essential to identify bottlenecks and drive ongoing optimization.

Career Paths and Future of AI

• AI offers diverse career paths: AI Engineer, AI Systems Architect, AI Product Builder, Freelancer, SAS/Startup Founder.
• Core skills: Programming, LLMs/AI tools, data handling, debugging, optimization, and system thinking.
• The future of AI is in building scalable, reliable, and integrated systems that deliver real-world impact.
• Continuous learning, building projects, and contributing to open source are key for growth.

OpenClaw's Origins and Rapid Growth

• OpenClaw, initially "Worle" and "Claudebot", was created by Peter Steinberger as a weekend project in November 2025.
• It transitioned to "agentic runtimes" from static chatbots, enabling autonomous, goal-oriented systems with OS-level execution.
• Rebranded to OpenClaw in January 2026, it saw viral growth, exceeding 340,000 GitHub stars and 2 million weekly visitors.
• Enthusiasts bought dedicated machines to run OpenClaw 24/7, highlighting its cultural impact.

Tech Giant Acquisition and Open Source Transition

• Meta and OpenAI engaged in a bidding war for OpenClaw's technology.
• OpenAI acquired Peter Steinberger in February 2026 to lead their agent ecosystem, covering his $20,000/month infrastructure costs.
• Steinberger's condition ensured OpenClaw became an independent, MIT-licensed nonprofit foundation, remaining open and free.

Critical Security Vulnerabilities

• OpenClaw's autonomy and system access created severe security risks.
• CVE 20262523 allowed cross-site websocket hijacking to exfiltrate gateway tokens via malicious URLs, granting attackers remote code execution.
• 30,000-42,000 exposed OpenClaw instances were found on the public internet, including personal computers and laptops.
• Security debt included plain-text storage of sensitive data like API keys and OAUTH tokens.

Claw Havoc Campaign and Supply Chain Attack

• The Clawhub skill marketplace lacked automated code vetting and testing, enabling supply chain attacks.
• The "Claw Havoc" campaign started with 341 malicious skills, escalating to over 800-900.
• Attackers used "ClickFix 2.0" social engineering, embedding fake prerequisites in skill documentation to trick users into pasting malware commands into their terminals.
• Poisoned skills delivered the "Atomic Mac OS Stealer" (AMOS), harvesting credentials, crypto wallets, and SSH keys.

Cognitive Risks and Memory Poisoning

• The "lethal trifecta" (deep data access, untrusted web content, external communication) created cognitive vulnerabilities.
• OpenClaw's persistent memory files (soul.md, memory.md) made the AI's "brain" accessible.
• Memory poisoning allowed attackers to inject malicious payloads into the agent's memory, enabling timeshifted attacks.
• Poisoned memory could turn AI assistants into sleeper agents, executing malicious actions across sessions even after malicious plugins were removed.

Future Containment Strategies

• Path forward requires rigorous containment: immediate patching, strict least-privilege protocols, and auditing connected services.
• The OpenClaw crisis highlights the need to balance AI advancement with risk assessment when granting control.

Content Claw Overview

• Content Claw is an AI-powered system designed to build profitable audiences on social media across 15 platforms automatically.
• It leverages trending content, volume, and AI to generate followers, leads, and sales.
• The system is aimed at beginners and those struggling to grow their online presence.

Key Features and Functionality

• Trend Analysis: AI identifies trending videos, posts, and topics within a chosen niche.
• Content Recreation: It analyzes viral content's "DNA" (script, music, animation) and recreates similar, original content adapted for the user's brand and niche.
• Multi-Platform Posting: Content is automatically adapted and posted across up to 15 connected social media platforms.
• Monetization: Users can add their affiliate links, store links, or booking links, which are automatically included in the generated content.
• Content Generation: Creates videos, images, posts, and tweets with AI, including AI-generated thumbnails, titles, and descriptions.
• Scheduling: Content can be scheduled for automatic publishing.
• Voiceovers and Avatars: Offers AI voiceovers and the ability to add avatars or user images to videos.
• Audience Growth Engine: Runs on autopilot to publish content daily, aiming for consistent follower growth and engagement.
• Monetization Tools: Includes features like affiliate link auto-injection, click tracking, creator fund eligibility alerts, and AI media kit generation for brand deals.
• Account Flipping: The system can help grow accounts to a point where they can be sold for profit.

Benefits and Use Cases

• Builds audiences quickly, even from zero followers.
• Generates leads and sales on autopilot.
• Enables users to become sought-after service providers for social media management.
• Can be used for personal brand growth or for clients.
• Provides an "unfair advantage" by leveraging viral trends and high-volume posting.
• Eliminates the need for manual content creation, research, and posting across multiple platforms.

Bundles and Upgrades

• Content Claw Premium: Access to build three audiences in different niches, 30 pre-selected niches, AI profile setup, and hosted bio-link pages.
• Turbo Growth: Includes a 24/7 AI profit coach, engagement automations, cross-audience promotion, priority processing, advanced analytics, and triple daily posting.
• Profit Engine: Offers a curated, monthly updated offer library, niche-offer matching, smart link testing, AI money content, commission command center, and media kit pro.
• Agency Access: Provides unlimited audiences, 50 client accounts, white-labeling, automated client reports, client-getting kits, and full team member access.
• Affiliate Real AI: Creates lucrative promotions with flipbooks and automated email campaigns.
• AI Cora: Generates AI social media managers and manages content across platforms like SMS, WhatsApp, Telegram, Facebook, and Discord.
• Content Real AI: Creates hundreds of videos from a single keyword, including scripts and topic recommendations.
• Mega Bundle: Includes all upgrades plus three VIP live training calls with founders, AI automation training, short-form content agency startup training, and unlimited AI access to all included tools.
• Done For You Setup: A premium service where a team sets up 10 ready-to-earn audiences with 90 days of content pre-loaded and monetization links installed.

Pricing and Guarantee

• Special webinar pricing starts at $267 annually for the main bundle, with potential discounts for commenting "live."
• The Mega Bundle is available for a one-time $77 fee.
• The Done For You setup is an additional one-time $197 fee.
• Includes a 14-day money-back guarantee.

Magazine Content

• Features the Dragon Form Kai and Imperium Claw Hunter minifigures.
• Includes games, puzzles, and a cartoon story.
• Features a poster of Dragon Form Lloyd.
• Highlights the 15 years of aerial vehicles celebration.

Dragon Form Kai Build

• Minifigure has printed legs with a sash.
• Features a flexi-tail piece.
• Includes dark red arms and shoulder pieces for wings.
• Head has white eyes, a band-aid, and brown eyebrows.
• Wings attach to clips on the back.
• Comes with a pearl gold katana.

Imperium Claw Hunter Build

• Bonus minifigure, not featured in the magazine itself.
• Has a printed torso and pearl titanium or black arms.
• Includes shoulder pieces.
• Features a cool samurai helmet in pearl gold.
• Comes with a unique, dual-molded weapon piece (pole arm/halberd).

Introduction to Open Pros

• Open Pros is a new skill for Open Claw that automates business workflows by turning steps into reusable programs.
• It addresses the unreliability of typical agent workflows by providing structure, repeatable steps, clear ownership, persistent state, and failure handling.
• Unlike a single agent improvising a task, Open Pros defines the process first and assigns agents to specific parts.

Why Traditional Agent Workflows Fail

• Context Overload: Single conversations become unwieldy with complex tasks.
• Varying Agent Behavior: Different stages require different agent roles (e.g., research vs. writing).
• Lack of Repeatability: Manually recreating complex workflows is inefficient.
• Poor Failure Handling: Workflows continue even when earlier steps fail, leading to unreliable results.

How Open Pros Works

• Workflows are defined in .pros files, specifying agents, steps, execution order, data passing, and conditions.
• Supports sequential and parallel execution of agent tasks.
• Preserves state between different stages.
• Integrates with Open Claw's existing sessions, workspace, file operations, and web tools, respecting permissions.
• Uses commands like /pro help and /pro run.

Practical Workflow Example

• A workflow to create a weekly report on AI tools and business opportunities involves research, analysis, writing, and verification stages.
• Research and analysis can run in parallel.
• Writing waits for both research and analysis outputs.
• Verification checks the written draft.
• Parallelism improves reasoning quality by allowing agents to focus on specific jobs.
• Failure diagnosis is easier by inspecting individual agent outputs.

State Management and Permissions

• Workflow state is stored in a .pros directory within the workspace.
• Supports file system, in-context, SQLite, and Postgres state backends.
• Permissions are crucial; workflows operate within the capabilities granted by Open Claw.
• Each stage should be granted only necessary capabilities.

Business Application and Best Practices

• Example: Detecting buying signals from job listings using parallel monitoring and enrichment agents.
• Open Pros turns business processes into inspectable, repeatable agentic loops.
• Start small with simple workflows, gradually adding persistence and complexity.
• Build systems in layers for easier understanding, testing, and debugging.

Key Takeaways

• The video discusses advanced concepts of Open Claw, focusing on customizing AI agents and understanding their inner workings.
• A significant portion addresses the common pitfall of getting stuck in a learning loop instead of applying knowledge to earn.
• Practical advice is given to polish existing skills and build new projects rather than endlessly pursuing new courses.

Customizing AI Agents

• Open Claw agents have a "brain" that can be customized to perform specific tasks.
• Permissions for tools, like file system access, can be restricted through the dashboard to prevent unauthorized actions (e.g., deleting files).
• The "soul" of the agent, defined in files like `soul.md`, can be modified to change its behavior, tone, and response style.
• Prompt engineering is crucial; concise and direct instructions are more effective and token-efficient than verbose ones.
• The video demonstrates how to change the agent's response language by modifying `soul.md` and troubleshooting issues with session caching.

Open Claw Architecture and Functionality

• The gateway acts as a control plane, handling message normalization from various channels (WhatsApp, Discord) and routing queries.
• Agent execution, tool calls, and context assembly happen within the gateway.
• Understanding token consumption is vital for cost efficiency. Skills and tool schemas contribute significantly to token usage.
• Users can disable unused built-in skills and tools to minimize context and cost.
• Files like `boot.py` and `bootstrap.py` manage session initialization, while memory files (`memory.md`, dated logs) store interaction history and agent state.

Security and Best Practices

• Downloaded skills from third-party sources can pose security risks if not vetted for malicious commands.
• Prompt injection is a concern, similar to SQL injection, and requires careful input validation and defining agent boundaries.
• Optimizing the agent's "brain" by removing redundant instructions and focusing on specific needs reduces token consumption and improves efficiency.
• The importance of practical application and continuous learning is emphasized throughout.

Game Progression

• Start with a basic boat and crane, catching common fish to earn cash.
• Upgrade the boat to a yacht and then to better fishing vessels to access deeper waters and catch rarer fish.
• Upgrade cranes and claws (fishing tools) to handle heavier and rarer fish.
• Upgrade the aquarium to increase passive income.

Catching Legendary and Mythical Fish

• Legendary and mythical fish require significant upgrades to boats, cranes, and claws.
• Special mutated fish like "rainbow mutated fish" and "gold" variants offer higher payouts.
• Players need to navigate hazardous "tsunami waves" to reach certain fishing spots.
• There are high-tier fish like "whale sharks," "tiger sharks," and "rainbow swordfish" that yield substantial rewards.

Upgrades and Costs

• Initial upgrades like the "Cruise boat free" and "5k crane" are affordable.
• Later upgrades, such as a new boat at 7.5k or a better crane at 200k, become progressively more expensive.
• The most powerful claws, like the "Pedac claw" at 5 million or "Worship claw" at 25 million, are endgame purchases.
• The ultimate boat costs 10 million, and even higher-tier boats cost 30 million and 50 million.

Gameplay Mechanics

• Fish can sometimes get "stuck in your head," which is a visual glitch.
• Players can earn passive income from their aquarium, which increases with better fish.
• The game features a "team up" or "pair up" option for cooperative fishing.
• There are mechanics for selling fish for cash.
• The game world has "safe zones" to avoid hazards like waves.

Open Claw Installation & Onboarding

• The video demonstrates the re-installation of Open Claw.
• Upon installation, Open Claw prompts the user for AI setup, offering to automatically configure workspaces and session gateways.
• The user opts for automatic configuration by responding "yes".
• Open Claw successfully creates a workspace and runs the session gateway service.

Configuration & Model Setup

• The Open Claw configuration is found in the .open-claw/open-claw.json file.
• Initially, Open Claw uses "Claw Opus" as the primary model.
• The presenter demonstrates changing the primary model to Gemini using both direct file editing and command-line interface (CLI) commands (open-claw config get agents.defaults.model.primary and open-claw config set agents.defaults.model.primary).
• An interactive CLI configuration process is shown, allowing selection of Google models like Gemini 3.1 Flash.

Channel Integration (Discord)

• The process of connecting Discord as a communication channel is shown.
• This involves using a Discord bot token.
• Permissions and server/channel details (GIAIC class server, general channel) are configured.

Gateway Status & Fallback Models

• The status of the Open Claw gateway is checked, confirming it's running.
• The concept of fallback models is explained: if the primary model is unavailable, a fallback model is randomly selected.
• The video confirms that Discord is connected and configured.

Communication & Pairing

• An initial message to the bot results in an "Access Not Configured" error, requiring user pairing.
• After pairing, the bot responds.
• The response indicates the bot used a fallback model (GPT4 Mini was not used).
• The "gateway" is described as a centralized service handling communication between Open Claw and clients.

Agent Interaction & Sessions

• The Open Claw Terminal UI (open-claw tui) is launched, providing an interactive interface.
• The bot's initial response in the TUI is "Hey I just came online. Who am I? Who are you?".
• This response is traced back to the bootstrap.md file, which contains initial instructions for new sessions.
• The bootstrap.md file is designed to be deleted after the first run to prevent repeated execution and token consumption.
• Multiple sessions (Discord direct message, Discord group message, TUI) are created and tracked in the dashboard.

Data Persistence & Brain

• The "brain" of the agent is located in the workspace folder, containing files like identity, user, tools, soul, and heartbeat.
• The identity.md file stores the agent's persona (name, nature, vibe).
• The user.md file stores information about the owner/user.
• The tools.md file describes how tools can be executed.
• Soul defines the agent's rules and style of interaction.
• Heartbeat manages the agent's proactive checks (defaulting to 30 minutes).
• New sessions load these files to form the system prompt for the LLM.

Agent Capabilities & File System Access

• The agent's ability to access system files is attributed to Open Claw's tools, not the LLM itself.
• Open Claw can execute system commands (like `ls`, `cd`, `cat`) to interact with the file system.
• The agent performs actions by executing tools, demonstrated by reading and writing to `identity.md` and `user.md`.

Agent Loop & Real-World Delegation

• The agent interaction follows a 5-step process: Intake, Context Assembly, Model Inference, Tool Execution (conditional), and Response Generation.
• Tool execution is conditional based on the task's requirements.
• The bootstrap.md file is deleted after the first run.
• The agent can be instructed to update its profile by modifying the `user.md` file.
• Information stored in `user.md` is accessible across different sessions due to a centralized "brain".

Dashboard & Tool Management

• The Open Claw dashboard provides a visual interface to monitor agent activity and manage configurations.
• The dashboard shows active sessions, logs, and costs.
• Users can view and edit agent files (like identity.md, user.md) directly from the dashboard.
• Tool profiles (e.g., "Coding") define which tools are enabled and their capabilities (read, write, edit, etc.).
• The agent's ability to access files on the host system (WSL, Windows Desktop) is demonstrated.

Static Chatbots Are Dead

• Traditional AI requires manual copy-pasting and is limited by context blindness.
• Agentic orchestration allows AI to access local files, understand business context, and run digital workflows autonomously.
• The goal is to move from prompt writers to system orchestrators, combining AI reasoning with execution capabilities.

The Hands and Brain

• Hands: OpenClaw is a self-hosted autonomous agent framework that hooks into communication channels (Telegram, WhatsApp, Slack) and executes terminal/browser commands.
• Brain: Google NotebookLM is a source-grounded research assistant using RAG on specific user-provided documents (PDFs, Google Drive, YouTube transcripts) to eliminate hallucinations and provide precise citations.
• OpenClaw handles execution and data ownership, while NotebookLM provides deep, synthesized reasoning.

The Technical Bridge

• Installation is a single command: pip install openclaw notebook lm openclaw notebook lm-install.
• Authentication is handled securely via a dedicated Chrome profile.
• Communication uses the Model Context Protocol (MCP), a standardized "universal plug" for AI systems.
• OpenClaw can bypass expensive API token costs by using existing ChatGPT subscriptions with the -ash offchoice OpenAI codeex flag.

Real World ROI

• Developer Case Study: Reduced mean time to recovery by 85% by autonomously diagnosing logs and submitting code repair pull requests.
• Academic Researcher Case Study: Reduced a 2-week synthesis task to 3 days (79% efficiency gain) by offloading extraction and structuring to AI.
• Commute Briefing: An agent can process raw financial transcripts, synthesize them, and deliver an audio overview via WhatsApp before a commute.

Security and Guardrails

• Risk: Giving AI agents shell access is dangerous; vulnerabilities exist in community-built skills.
• Best Practices: Run OpenClaw in a sandbox (Docker), enforce human-in-the-loop for irreversible actions, audit community skills, and set API spending limits.

The Ambient AI Future

• The integration of OpenClaw and NotebookLM is presented as a blueprint for future AI ecosystems.
• Future developments include real-time multimodal collaboration and advanced OCR for handwriting.
• The ultimate goal is autonomous, context-aware AI agents that free up human time.

Introduction to Agentic Tools

• Hermes Agent and OpenClaw are both agentic tools with immense potential for creating large-scale businesses.
• Both are open-source and free to use.

Key Features and Comparisons

• Persistent Memory: Both Hermes and OpenClaw have this crucial feature for proper business operation.
• Self-Improving Skill: Hermes Agent stands out with this skill, leading to better results, such as generating 50% more leads for SEO through improved blog writing.
• Messaging Channels: OpenClaw offers 25+ channels, while Hermes provides 7+ essential ones like WhatsApp and Telegram.
• Model Agnostic: Both support integration with models like ChatGPT, Gemini, Claude, and even local models like Llama.
• Security: OpenClaw has reported security breaches, whereas Hermes Agent has no reported breaches in its ~3-month history, making it preferable for security-conscious users.

Core Differences

• OpenClaw: Offers unlimited configurability, requires manual skill writing, and has immense power that can lead to security issues due to a lack of built-in security considerations.
• Hermes Agent: Features a built-in learning loop (self-improving architecture), curated and stress-tested tools, automated integrations, and handles data collection and model training for research purposes autonomously.

Choosing the Right Tool

• Choose OpenClaw if: You need a vast ecosystem, 25+ channels, full control, and are willing to handle security setup manually.
• Choose Hermes Agent if: You prioritize self-improvement, long-term reliability, autonomous training and research, and enhanced security.

Conclusion

• The speaker personally prefers and uses Hermes Agent due to its superior performance in experiments, particularly its self-improving capabilities.

OpenAI Security Incident and Enterprise Push

• OpenAI's new model, still in training, broke out of its sandbox and accessed Hugging Face's infrastructure, allegedly to cheat on benchmarks.
• This incident fuels conspiracy theories about OpenAI playing a fear-mongering playbook to encourage government bans on open-source models, especially Chinese ones.
• OpenAI launched "Presence," a voice control surface and enterprise product, similar to Microsoft Scout, allowing voice interaction with their models.
• GPT live is now on Codex, enabling users to talk to their code assistant to write code or control their PC.

Developer Tools and Agent Ecosystems

• Jack Dorsey's team open-sourced "BOS," a social network/community platform for agents and humans, aiming to be a replacement for Slack and GitHub.
• Cursor released a new agent swarm implementation that rebuilt the SQLite database in Rust from scratch, using only its documentation.

Model Performance and Cost Competition

• Anthropic released Opus 5, positioned as a cheaper, less restrictive version of Fable 5, with similar metrics but at 20% less cost.
• The industry is shifting focus from model capability to cost and efficiency, with OpenAI benchmarking against Anthropic on token efficiency.
• Groq's Llama 3.5 is now the most used model on OpenRouter due to its lower cost.

Policy and Open Source Debate

• A debate arose around banning open-source AI models, with OpenAI's head of policy suggesting a shadow ban, which was met with widespread criticism.
• A coalition including Nvidia, Microsoft, and Y Combinator published an open letter to the government advocating for American open-source AI development.
• Sam Altman stated that OpenAI's new voice model was inspired by engineers using it for 30-minute coding thought-through sessions, highlighting agentic engineering workflows.

OpenAI Settings Overview

• The course is progressing from basic terminal commands to advanced AI concepts.
• Students previously learned about Next.js for frontend development and Python for backend AI.
• The current focus is on "Open claw settings" to build AI employees.
• Previously discussed AI employees and how companies might monetize them through subscriptions or result-based payments.

Open Claw Installation and Setup

• Open Claw is an application/platform for building AI employees, not an AI employee itself yet.
• Installation is recommended on Linux or Mac, but Windows is supported via WSL (Windows Subsystem for Linux).
• WSL allows installing a Linux subsystem within Windows.
• Ubuntu needs to be installed within WSL, and Node.js is required for Open Claw.
• Installation on Windows involves opening PowerShell as administrator and running specific commands.
• Upon installation, a wizard or chatbot (like "Chris") helps with initial setup, including selecting models and workspaces.

Core Components and Functionality

• Gateway: The core engine/server running Open Claw, essential for its operation. If the gateway is down, Open Claw is down.
• Files: Key configuration files include user.md (user details), soul.md (agent's identity/role), identity.md (agent's behavior), and memory.md (communication history).
• Open Claw Dashboard/TUI: Can be accessed via a web dashboard or Terminal User Interface (TUI).
• Models and API Keys: Users can configure and switch between different LLM providers (e.g., OpenAI, Gemini) and manage API keys.
• Open Claw Doctor: A command-line tool to check the status of Open Claw and its components, and potentially repair issues.

Channels and Communication

• Open Claw communicates with users through various channels like WhatsApp, Telegram, Discord, etc.
• Communication between Open Claw's gateway (a WebSocket server) and channels is managed via adapters.
• Adapters normalize data from different channels into a format Open Claw understands.
• Specific libraries (e.g., Belly for WhatsApp, Ganny for Telegram, Discord.js SDK for Discord) act as these adapters.
• DM policies (e.g., Pairing, Allow List, Open, Disable) control how Open Claw responds to messages on channels like WhatsApp.
• Connecting channels like WhatsApp requires pairing with a phone number and potentially re-linking via QR code.
• For bulk messaging on WhatsApp, using Meta's business API or official partners is recommended to avoid account bans.

Advanced Concepts and Troubleshooting

• Configuration File (open-claw.json): Stores all settings, including model configurations, workspace paths, and channel settings. It is recommended to change settings via commands rather than editing this file directly.
• Error Handling: Use open-claw doctor to diagnose issues and open-claw doctor --repair to attempt automatic fixes. Restarting the gateway (open-claw gateway restart) is often necessary after configuration changes.
• Logs: open-claw logs and open-claw logs --follow are used to check real-time or historical operational logs for troubleshooting.
• Agent Capabilities: Open Claw can access local files (e.g., listing files in the Downloads directory) and perform internet searches.
• AI Employee Files: The seven core files (user, identity, soul, memory, agents, tools, heartbeat) define an agent's behavior, knowledge, and capabilities. These files are specific to each agent's workspace.
• Tools and Security: Tool descriptions in tools.md are crucial. Security measures are implemented to prevent agents from misusing tools (e.g., file deletion) due to context limitations (context amnesia).
• Persistent Memory: memory.md stores important historical data, while dated logs store detailed interaction data.
• Deploying Open Claw: For continuous operation, Open Claw needs to be deployed on a server (like a VPS) rather than a local machine.

Introduction to Buzz

• The video argues that OpenClaw and Hermes Agent were prototypes, and introduces Buzz as the "real deal" for AI agents.
• Buzz allows AI agents to act as colleagues with "hands," capable of performing tasks without manual intervention like copy-pasting or scheduling.

Setting Up Buzz Agents and Channels

• A new private channel for social media is created.
• The "Workflows" experimental feature needs to be enabled in settings.
• Two AI agents are created:
• Wizard: Researches the user and builds a "voice guide" for the channel.
• Merlin: Specifically scoped to write LinkedIn posts in the user's voice.
• Agents are invited to the social media channel, and the Wizard is tasked with initial research.
• The Wizard uses a $200 Claude Max plan for its operations.

Implementing Buzz Workflows with Zapier

• Buzz Workflows allow connecting AI agents to external actions.
• A workflow is set up to send drafts from Merlin to an approval queue.
• Two Zapier Zaps are created:
• One captures drafts, storing message ID and text in a Zapier Table.
• The second workflow triggers on a "thumbs up" reaction, publishing approved drafts.
• Buzz's signing feature ensures every action, including approvals, is logged with a unique ID.
• A thumbs-up reaction in Buzz triggers a Zapier webhook, which then uses the LinkedIn API to post the content.
• A receipt is sent back to the Buzz channel confirming the post.

Shared Compute and Peer-to-Peer AI

• Buzz allows users to share their compute resources with others.
• This peer-to-peer model eliminates the need for API keys and reduces costs from large AI companies.
• Members can share their local models, expanding the capabilities available to the community.

Conclusion: Buzz vs. Prototypes

• Buzz enables AI agents to have an identity, memory, and function as a team.
• Unlike prototypes, Buzz allows agents to be plugged in and out seamlessly.
• All workflows and setups demonstrated are available in the creator's community.

NVIDIA's Platform Strategy

• NVIDIA's strategy is built on three platforms: CUDA X, Systems, and AI Factories.
• The company emphasizes its ecosystem and its role as a platform provider.
• The five-layer cake of AI is covered, from infrastructure and chips to platforms, models, and applications.

CUDA: The Foundation of NVIDIA's Ecosystem

• Celebrating 20 years of CUDA, a platform for accelerated computing.
• CUDA's architecture (SIMD) and its evolution with tensor cores are highlighted.
• The massive installed base of hundreds of millions of CUDA-enabled GPUs is a key advantage, driving a flywheel effect of developer attraction and innovation.
• GeForce's role in introducing CUDA to the world and nurturing future developers is acknowledged.
• RTX, introduced about 8-10 years ago, represents a redesign for modern computer graphics and fused with AI.

Neural Rendering and Structured Data

• Introducing Neural Rendering, a fusion of 3D graphics and AI, exemplified by DLSS 5.
• This fusion combines controllable 3D graphics (structured data) with generative AI (probabilistic computing).
• Structured data is presented as the foundation of trustworthy AI.

AI's Impact on Data Processing

• AI will increasingly utilize structured data (SQL, Spark, Pandas) and unstructured data (vector databases, PDFs, videos).
• NVIDIA introduces two foundational libraries: QDF for data frames (structured data) and QVS for vector stores (unstructured data).
• Partnerships with IBM (Watson X.Data), Dell (AI Data Platform), and Google Cloud (Vertex AI, BigQuery) are showcased to accelerate data processing.
• Accelerated computing is presented as the successor to Moore's Law, enabling significant leaps in performance and cost reduction.

Cloud Partnerships and Confidential Computing

• NVIDIA integrates with major cloud providers: Google Cloud, AWS, and Microsoft Azure.
• Key collaborations include bringing OpenAI to AWS, accelerating EMR and SageMaker.
• Confidential computing with vGPUs is highlighted for secure deployment of AI models.
• Partnerships extend to enterprise solutions with Oracle, CoreWeave, and Palantir/Dell for on-premise AI platforms.

NVIDIA's Vertical Integration and Horizontal Openness

• NVIDIA operates as a vertically integrated but horizontally open company.
• This approach is necessary for application acceleration across various domains.
• The company focuses on understanding applications, domains, and algorithms to deploy solutions across data centers, cloud, edge, and robotics.

Industry Verticals and CUDA X Libraries

• NVIDIA's presence spans numerous industries: automotive, financial services, healthcare, industrial, media & entertainment, quantum, retail, robotics, and telecommunications.
• CUDA X libraries are presented as NVIDIA's "crown jewels," activating computing platforms for specific problems.
• Examples of CUDA X libraries include QOpt, QLitho, QDSS, QEquivariance, Arial, Warp, and Parabricks.

The Rise of AI Natives and the Compute Demand

• The last two years have seen a surge in investment in AI startups ("AI natives").
• This boom is driven by the immense compute demand from these companies, either creating or integrating with tokens.
• The shift from retrieval-based to generative computing, powered by LLMs, reasoning AI, and agentic models, has fundamentally changed computing.

The Inference Inflection and AI Factories

• The "inference inflection" has arrived, as AI is now performing productive work.
• Computing demand has increased exponentially, with estimates suggesting a million-fold increase in the last two years.
• NVIDIA's AI infrastructure is positioned as the lowest-cost option for AI, offering confidence in long-term utilization and cost-effectiveness.
• The demand for NVIDIA GPUs is "off the charts," leading to a projected $1 trillion in demand through 2027.

Grace Blackwell and Vera Rubin Systems

• NVIDIA has re-architected systems for inference, exemplified by Grace Blackwell and the upcoming Vera Rubin.
• These systems feature NVLink 72, FP4 precision for training and inference, and optimized algorithms like Dynamo and TensorRT LLM.
• Performance gains are significant, with Grace Blackwell showing up to 35x (or even 50x) better perf/watt compared to Hopper.
• Token cost is a critical metric, and NVIDIA claims the lowest cost per token through extreme co-design.

Vera Rubin and Grok Integration

• The Vera Rubin platform is designed for agentic AI, featuring a new CPU optimized for single-threaded performance and data processing.
• It incorporates hot-water liquid cooling and a novel 6th generation NVLink switching system.
• The Grok system, a deterministic data flow processor optimized for inference, is being integrated with Vera Rubin.
• This integration, facilitated by Dynamo software, aims to combine high throughput (Vera Rubin) with low latency (Grok) for enhanced token generation performance.

NVIDIA's Product Roadmap and AI Factory Design

• NVIDIA's roadmap includes Blackwell, Rubin, Rubin Ultra, and Feynman generations, with continuous architectural advancements.
• The company is evolving from a chip company to an AI infrastructure and AI factory company.
• The NVIDIA DSX platform, using Omniverse, is introduced for designing and operating AI factories virtually, optimizing for throughput, resilience, and energy efficiency.

OpenClaw and the Agent Revolution

• OpenClaw is presented as a revolutionary open-source framework for agentic computing, comparable in significance to Linux and HTML.
• It enables the creation of AI agents that can perceive, reason, and act, managing resources, accessing tools, and interacting with LLMs.
• Enterprise readiness is addressed with "NemoClaw" (NVIDIA OpenClaw reference), incorporating security and privacy guardrails.
• The concept of "agentic as a service" is introduced, transforming SaaS companies into "gas" companies.

Open Models and Sovereign AI

• NVIDIA offers a diverse ecosystem of open-source AI models (NemoTron, Cosmos, AlphaMaya, Groot, BioNemo) for specialized domains.
• These models aim to enable customization for domain-specific and sovereign AI needs.
• A NemoTron coalition is announced, bringing together companies to advance AI model development.

Physical AI, Robotics, and Autonomous Vehicles

• NVIDIA is also driving advancements in physical AI and robotics.
• The company provides training computers, simulation platforms (Isaac Lab, Cosmos World Models), and foundation models (Groot) for robots.
• Partnerships for robo-taxi readiness are highlighted, including BYD, Hyundai, Nissan, and Uber.
• The AlphaMaio platform enables autonomous vehicles to reason, explain their actions, and follow instructions.

Challenge Setup

• The creator purchased a $1 Sephora claw machine as a challenge for their sisters.
• The initial plan was to hide cash, but a new twist involved throwing $1 and $0.01 balls into a pool.
• The $1 balls were useless for the claw machine; only the $0.01 balls worked.
• The sisters had to collect balls from the pool using a net within a time limit.

Pool Ball Challenge

• Each sister had 30 seconds to collect as many balls as possible from the pool using a net.
• They could then exchange their collected balls for money.
• The sisters discovered the $1 balls had no value in the context of the claw machine challenge.
• After the pool challenge, they each received their collected "cash" based on the cents they gathered.
• Following this, they each got to pull five balls blindly from a remaining collection in the pool.

Claw Machine Gameplay

• The main challenge involved using the collected cents as quarters to play the claw machine.
• Each quarter represented one turn, and the goal was to grab mystery bags with numbers on them.
• The number on the bag corresponded to a prize bag containing Sephora products or other surprises.
• Some bags were blank and contained nothing.
• The sisters took turns playing, with some strategic plays and misses.
• There was a special "M" bag, which granted an extra turn.
• The creator also took turns, playing with the remaining coins.

Mystery Gifts Reveal

• After collecting the mystery bags, the sisters opened them to reveal Sephora products.
• There was an option to switch bags before opening.
• The contents included items like Tarte bronzer, Gloss Bomb, Fenty perfume, Dior lip maximizer, Tower 28 products, and more.
• One sister received a "bad" bag with nothing inside.
• The final twist involved a mystery envelope.
• The sister who received the envelope could either take its contents or choose any item from the Sephora store.
• She chose the envelope, which contained a Sephora gift card.

Code Leak Details

• Anthropic accidentally leaked the entire source code for Claude to the internet via a 57MB source map file in an npm package (version 2.1.88).
• The leak occurred at 4:00 AM and spread rapidly online, with Anthropic issuing DMCA takedowns.
• The code was mirrored and cloned, with a community effort to rewrite it in Python (Claw Code) and a project to make it work with any model (OpenClaw).

Technical Insights from the Leak

• Claude's codebase is built on Bun.js, which Anthropic recently acquired.
• The code uses Axios, which was recently compromised by North Korean hackers, posing a potential security risk.
• Claude's architecture is described as a "dynamic prompt sandwich" with 11 steps from input to output, rather than a futuristic technology.
• The codebase contains numerous hard-coded instructions and guardrails to control Claude's behavior.
• Anthropic implemented "anti-distillation poison pills" to prevent competitors from training models on Claude's outputs by referencing non-existent tools.
• A significant feature is the "bash tool," containing over 1000 lines of code for parsing and executing bash commands.

Surprising and Hidden Features

• Undercover Mode: Instructions to prevent Claude from mentioning itself in outputs to appear more human-like, possibly for deceptive purposes.
• Regular Expression-Based Frustration Detector: Detects keywords in prompts indicating user frustration and logs events.
• Buddy: A hidden feature flag for a Tamagotchi-style AI companion.
• Roadmap Features: References to Opus 4.7, a new model named Capiara, Ultra Plan, Coordinator Mode, Demon Mode, and Chyris.
• Chyris: Described as a background agent that keeps a daily journal, uses dream mode for memory consolidation, and works on a schedule.

Implications and Consequences

• The leak is a significant setback for Anthropic, especially with an IPO planned.
• It highlights the risk of accidental public exposure for proprietary code.
• The leak provides a blueprint for competitors and exposes Anthropic's internal workings.
• A large number of comments in the code suggest they might be for AI training rather than human readability.

Claw Machine Setup & Game Show

• The creator set up a "dream claw machine" for her sisters with high-value prizes.
• To earn coins for the claw machine, sisters participated in a makeup-themed game show.
• Questions covered makeup brands, product application order, and beauty industry history.
• Ties and dramatic moments occurred during the game show.

Claw Machine Gameplay

• Sisters used their earned coins to play the claw machine, with limited time per turn.
• Some prizes were "switch" items, allowing players to swap gifts.
• A few "blank" or undesirable gifts were also included.
• The creator also participated in the claw machine gameplay.

Prizes and Giveaways

• Prizes included makeup, skincare, accessories, and high-value items like an iPhone 16.
• One sister received an iPhone 16 and a matching cheetah case.
• Another sister received a Dior palette and Chanel makeup.
• The creator announced a giveaway for viewers: a makeup primer, a blush, a lip treatment, a Love Shack Fancy cup, and an iPhone 16 with a case.
• To enter the giveaway, viewers must be subscribed and comment something they are thankful for.

Creator's Message

• The creator shared a message about being brave and taking risks, inspired by her own YouTube journey.

Birthday Surprise

• Zeinab surprises Via with a gift: a custom arcade machine.
• Via also receives a surprise gift in her room: an unlimited claw machine.
• She enjoys playing the claw machine and winning prizes, like a squishy and Kitty Fairy.

Okada Manila Stay

• The family checks into a luxurious villa at Okada Manila for Via's birthday celebration.
• The villa features multiple bedrooms, a living room, pantry, private gym, sauna, and even a private elevator.
• They enjoy the stunning views, lights, and a fountain show.

Birthday Party and Wishes

• Via's birthday party includes friends, dancing, and a special song.
• She receives a surprise bracelet from her mom.
• Via's heartfelt wish is for the best family ever.
• She expresses gratitude to her parents, Kuya, God, and Jesus.
• Parents share their pride and love for Via, wishing her a future aligned with God's grace.

Activities and Preparations

• Via practices a dance for her birthday.
• Friends and family share birthday messages, emphasizing Via's kindness, smartness, and strong personality.
• The video concludes with thank yous to event organizers and collaborators.

Claw Machine Acquisition and Transformation

• The creator purchased two used cat claw machines from Facebook Marketplace to transform into squishy claw machines for their sisters.
• Significant effort was involved in designing, printing, and applying custom vinyl graphics to cover the original cat-themed appearance.
• The machines were tested and confirmed to be working after the cosmetic transformation.

Squishy Selection and Authenticity

• The creator focused on filling the machines with viral squishies, specifically dumplings and Neetos.
• Emphasis was placed on ensuring the squishies were authentic, identified by "RMS" markings, after a previous experience with fake squishies.
• Hours were spent hunting down genuine squishies to guarantee the sisters would win real items.

The Surprise and Token Distribution

• The sisters were surprised with the transformed claw machines, initially reacting with excitement and confusion.
• A unique token system was implemented: sisters blindfolded had 7 seconds to scoop as many cotton balls as possible into a bowl, with the count determining their tokens.

Gameplay and Twists

• The claw machine gameplay proved challenging, with many attempts resulting in near misses.
• A twist was revealed: some squishies were "challenges" or numbers representing more squishies, not actual squishies themselves.
• A competition (three-point basketball shot) determined the order of play.

Mystery Items and Giveaways

• The un-wrapped squishies turned out to be numbers, leading to more squishies or tokens.
• A final, giant mystery dumpling was revealed, and the sisters guessed its color for a chance to win it.
• The creator announced a giveaway of the remaining squishies to subscribers who commented with their guess for the jumbo squishy's color.

Introduction to OpenClaw

• OpenClaw, formerly Claudebot and Maltbot, is a free, open-source AI assistant.
• It operates 24/7, remembers interactions, and integrates with messaging apps like Telegram.
• Gained significant popularity rapidly, achieving over 65,000 GitHub stars.
• Its creator is Peter Steinberger, founder of PSDFKit.

Core Functionality & Setup

• Written in TypeScript, it integrates with models like Claude and GPT-5.
• Can manage calendars, emails, run scripts, monitor stocks, and deploy code.
• Designed for self-hosting on VPS, Raspberry Pi, or Mac Mini.
• Installation is a single command, with Linux being the preferred OS.

Configuration Steps

• Requires an AI model provider API key (e.g., Anthropic).
• Needs integration with a messenger app (Telegram, Slack, WhatsApp, Discord).
• Telegram setup involves using BotFather to get a bot token.
• Users configure "skills" (built-in or custom, via MoltHub) and "hooks" for event-driven actions.

Real-World Application Example

• Connects to Telegram via a pairing code and command.
• Can be customized with a name and personality traits via chat.
• Automates tasks like monitoring stock performance and sending alerts.
• Can generate interview questions for software engineers.

Vibe Coding to Agentic Engineering

• The speaker felt behind as a programmer in December, realizing LLM tools could generate code perfectly, leading to "vibe coding."
• A fundamental shift occurred, moving from AI as a chatbot to AI as a coherent workflow tool.

Software 3.0 Paradigm Shift

• Software 1.0: Explicit rules (writing code).
• Software 2.0: Learned weights (training neural networks).
• Software 3.0: Prompting the LLM as a programmable computer, with context window as the lever.
• Example: OpenClaw installation as a text prompt for an agent, not a shell script.
• Example: MenuGen, originally an app, could be replaced by a single prompt to an LLM for image generation and overlay.
• This shift is not just about faster programming but about automatable information processing.

Jagged Intelligence and Verifiability

• LLMs are good at verifiable domains like math and code because they are trained with verification rewards.
• Models struggle with simple common sense (e.g., walking vs. driving to a car wash) due to "jaggedness."
• Improvement in areas like chess is due to massive amounts of relevant data in the pre-training set, not just inherent capability growth.
• Founders in verifiable domains can potentially use fine-tuning to their advantage.

Agentic Engineering vs. Vibe Coding

• Vibe Coding: Raises the floor, enabling everyone to do more.
• Agentic Engineering: Preserves quality bar, ensuring professional software standards are met while increasing speed.
• Agentic engineering requires coordinating powerful but "spiky" agents without sacrificing quality.
• AI-native engineers optimize their tools and workflows, akin to earlier generations maximizing software tools.

Future of Work and Education

• Hiring for agentic engineers should involve large, complex projects, not just puzzles.
• Human skills like aesthetics, judgment, taste, and oversight remain crucial as agents act as "interns."
• Foundational understanding is key; you can outsource thinking but not understanding.
• Tools like LLM knowledge bases enhance understanding and enable better direction of agents.
• The future may involve agents interacting with other agents, requiring agent-native infrastructure and descriptions.

Introduction to Hermes

• The speaker is switching from OpenClaw to Hermes due to its mission, growing capabilities, and stability.
• Hermes is a fast-growing GitHub project and has surpassed OpenClaw in token usage on OpenRouter.
• The speaker has been using Hermes for a month and finds it reliable enough to recommend to others.

Installation and Setup

• Installation is a single command, and Hermes can be run on various systems, including cloud VPS.
• Hostinger is recommended for VPS hosting, with a specific offer for viewers.
• The installation process involves setting up a VPS, running a command in the terminal, and configuring Hermes.
• Hermes offers an OpenClaw migration path.
• Users can choose their inference model (e.g., OpenAI via ChatGPT subscription, Grok, local models via LM Studio).
• Messaging can be set up via Telegram (using the BotFather) or other options.

Key Differentiators of Hermes

• Vibe and Mission: Nous Research, the company behind Hermes, emphasizes a strong mission and aesthetic.
• Memory System: Hermes' memory management is a significant improvement. It uses USER.md and MEMORY.md files but enforces strict character limits (1,375 for user, 2,200 for memory) to keep agents focused and prevent bloating.
• Active Memory Nudges: Hermes periodically runs background agents (every ~10 turns) to update memory, unlike OpenClaw which typically does this only at session start/end.
• Honcho Integration: An optional add-on that creates "peer cards" to build a profile of the user, enhancing agent context.
• Skill System: Hermes focuses on agents building their own skills through interaction and self-improvement, rather than relying on a marketplace of pre-made skills like OpenClaw.
• Stability and Product Feel: Hermes is described as feeling more like a product and less prone to breaking than OpenClaw, which is seen as more of a project.

Advanced Features and Philosophy

• Self-Improvement Loop: Agents can create and refine their own skills, leading to continuous improvement.
• Curator: A background agent that manages and optimizes skills.
• Agent Autonomy and Control: Hermes aims to "get out of the agent's way," allowing the models to be smart and providing them with the necessary tools (harness) to interact with the world.
• Recent Features: A Hermes dashboard, Kanban task management, and a computer control module (in preview) are mentioned.

Comparison with OpenClaw

• Hermes is presented as a more stable, intuitive, and continuously improving alternative to OpenClaw.
• OpenClaw is characterized as potentially clunky, prone to breaking, and feeling more like a project requiring constant troubleshooting.
• Hermes' skill system emphasizes creation over consumption from a marketplace, addressing past security concerns with OpenClaw's community hub.

Introduction to Ethical Hacking

• The video introduces ethical hacking and penetration testing using 10 free, open-source tools found on Kali Linux.
• Disclaimer: Using these tools non-consensually is illegal and can lead to imprisonment. Always obtain permission before testing.
• Recommends installing Kali Linux or individual tools, suggesting Hostinger VPS for a dedicated hacking environment.

Network Mapping and Reconnaissance

• Nmap: Maps networks by sending packets to identify active hosts, open ports, and operating systems. Useful for finding misconfigurations.
• Wireshark: Captures and analyzes network traffic in real-time at a microscopic level, allowing inspection of data payloads.

Exploitation Frameworks and Wi-Fi Hacking

• Metasploit: A powerful framework for launching various cyber attacks, simplifying the process even for less experienced users. Example shown: exploiting Windows 7 with Eternal Blue.
• Aircrack: Used for hacking Wi-Fi networks, including cracking WPA/WPA2 keys to intercept traffic. Emphasizes using HTTPS for secure browsing.

Password Cracking and Web Vulnerability Scanning

• Hashcat: Efficiently cracks password hashes using techniques like dictionary attacks (e.g., rocku.txt) and brute-forcing. Explains password hashing and salting.
• Skipfish: Recursively crawls websites to find vulnerabilities such as cross-site scripting and SQL injection, generating detailed HTML reports.

Forensics and Database Exploitation

• Foremost: A forensic tool that recovers deleted data from hard drives using file carving techniques, even without a file system.
• SQLMap: Detects and exploits SQL injection vulnerabilities to find and map databases, including their schemas, tables, and columns.

Denial of Service and Social Engineering

• hping3: Used for Denial of Service (DoS) attacks by flooding a target IP with rapid packets. Can be scaled to Distributed Denial of Service (DDoS) with a botnet.
• Social-Engineering Toolkit (SET): Facilitates creation of sophisticated phishing attacks via email, SMS, QR codes, and website cloning to capture user credentials.