The video compares Qwen 3.8, GPT-5.6 Soul, Fable 5, and Kimi K3 across 23 side-by-side builds on GoldyBench, focusing on game and interactive demo quality such as Outrun, Skyrim-like RPGs, a flight simulator, GTA-style gameplay, Doom-style shooters, physics tests, and landing-page design. The host notes Qwen 3.8 is a massive 2.4T-parameter model and, in these tests, often produces smooth, fun, highly detailed outputs, sometimes with weaker UI, lighting, or buggy controls, while Kimi K3 frequently excels in front-end feel and physics interactions. Fable 5 remains the host’s favorite overall, with Kimi K3 generally second, and GPT-5.6 Soul often competitive but inconsistent. The takeaway is results are a mixed bag, so viewers should test models themselves, with links to GoldyBench and the AI Profit Ballroom community and trainings.

Discover how to master the Claude Agent OS to automate your business with custom design skills and voice agents. Learn practical strategies for SEO ranking, building agent memory, and scaling your AI workflows without hitting usage limits.

The script covers Alibaba’s announcement of Qwen 3.8 Max Preview, described as a 2.4 trillion parameter open source model available via Alibaba’s Token Plan (Coder/Coder Work) and Qwen Cloud, though the narrator can’t access it yet or find an OpenRouter API and notes it’s available in China. Qwen claims it is “second only to Fable 5,” which the narrator calls unusually bold, especially after China’s Kimi K3 released this week and performed extremely well in the narrator’s personal tests across 50 builds on Goldy Bench. The narrator argues open source Chinese models are rapidly closing gaps with and sometimes overtaking US frontier models, with more releases expected from GLM, DeepSeek GA4 Pro, and Minimax, and promotes AI Profit Boarding for training, community support, and future Qwen testing via Agent OS.

The video shows how to access and use Kimi K3 for free at kimi.com by signing in with a free account, switching the default model from K2.6 to K3 (or K3 Swarm), and choosing settings that conserve tokens. The creator recommends avoiding K3 Swarm on a free plan due to token limits, keeping the context window on standard (extra-long is premium), and adjusting thinking effort (standard/high/max) since higher effort uses more tokens. A quick example prompt demonstrates starting a coding task, and the video notes K2.6 can be faster for lightweight requests. It also highlights API testing results, examples of games/3D worlds and video generation using ReMotion, mentions free usage may be paused during peak times, and says Kimi K3 will be open source from July 2026 but requires powerful hardware to run locally.

The episode covers Qwen 3.8’s newly announced launch, described as an open-source model going live soon but currently only usable for coding in mainland China, with the international site not yet offering access. The host highlights its reported 2.4 trillion parameters and Qwen’s claim that it’s among the most powerful models available, second only to Fable 5 and said to beat models like GPT 5.6 and Kimi K3. The video contrasts restricted access to models such as GPT 5.6 and Fable 5 with free, open-source Chinese models like Kimi, which the host says performs exceptionally well in Goldy Bench tests and coding benchmarks. It also discusses Dean W. Ball’s comments on Kimi K3, open-weight risks, and possible future US regulatory pressure on open models.

The episode reviews Alibaba’s Qwen 3.8 (2.4T parameters) and its claim of being second only to Fable 5, then tests it side by side against Fable 5 and GPT-5.6 across multiple frontend/game-style builds (racing, fireworks, aurora, black hole, neon blaster/racer, cloth simulation, flight sim, promo video, RPG/GTA/Doom, Dragonflight, and Nordic crypt). The creator finds Qwen 3.8 often strong at fun 3D gameplay but frequently weaker in UI/graphics polish versus GPT-5.6 and Fable 5, with mixed results depending on the test. They note Qwen 3.8 is a big step up from Qwen 3.7 but may disappoint relative to its marketing, and they generally prefer Kimi K3 for front-end quality, while also discussing access difficulties and recommending using Coder/Qoda to try Qwen 3.8.

Discover how to combine Obsidian and NotebookLM to create an 'Infinite Context Engine' that gives your AI agents a permanent memory. Learn the workflow for capturing, organizing, and deploying your knowledge so your AI gets smarter with every interaction.

Learn how to access Claude Code and over 70 AI models for free using Omniroot's open-source gateway. This tutorial shows you how to set up an automated coding environment with auto-fallback to bypass rate limits and secure billions of free tokens for your projects.

Discover the ultimate AI Agent OS designed to automate content creation, coding, and research using tools like Claude and Notebook LM. Learn how to build your own integrated AI system for free to streamline your business workflow and boost productivity.

The host answers recent AI Profit Boardroom questions, including how to sync an Agent Operating System across five Macs by using Google Workspace/Drive with a shared API access token or running all Agent OS instances on a VPS and saving files locally (with tutorials using Cloudflare and Hostinger). He explains the latest ChatGPT app where Codex is rebranded to ChatGPT Work, combining chat and project-building features and supporting tasks like building apps, websites, videos, integrations, and scheduled tasks. He covers tools and training for creating AI avatar videos for training courses, citing high Twitter impressions and community examples. He advises how to use the free Omniroute/Omniroot coding API with various agents and troubleshoot “fetch failed,” and shares an Omi alternative for Windows that can sync with Obsidian via plugins.

The script reviews the release of GPT 5.6 and showcases 50 real-world builds created primarily with the flagship Sol model, including smooth web games (Doom-style, 3D racing, Space Invaders, pool physics, raycaster maze, RPG) and polished websites and dashboards. It explains the three new GPT 5.6 models—Sol (flagship), Terra (day-to-day workhorse), and Luna (fast, cheap, token-efficient)—and suggests when to use each, including a slower Pro option. It demonstrates using GPT 5.6 inside Codex and Agent OS via OAuth, highlights autonomous ��/goal” mode in a sandbox, and notes image generation and workspace organization. It also describes video creation via a Remotion-based video agent and using GPT with the Hermes system (web search, images, voice). Finally, it compares Sol with Fable 5 on cost and quality and promotes getting the Agent OS setup through the AI Profit Boardroom.

00:00 GPT 5.6 First Look
00:22 Showcase Builds
00:41 Sol Terra Luna Explained
01:51 Best Setup With Codex
02:16 Goal Mode Autonomy
03:12 Agent OS Workflow
04:01 Video Agent Pipeline
04:23 Hermes Agents Overview
05:23 Goals Board Productivity
05:54 More Game Demos
07:32 Luna Terra Comparison
08:37 Versus Fable 5 Pricing
09:21 Codex Versus Claude UX
10:08 Get Agent OS Access
11:16 Final Thoughts

The video compares OpenAI’s flagship GPT-5.6 Sol against Anthropic’s Claude Fable 5 in side-by-side tests, focusing on game generations and coded outputs. Sol is described as cheaper on output price and often smoother, less laggy, and strong on action and precision, winning rounds like an open-world dragon game, Dragon Realm, and narrowly edging a 3D racer on graphics, while Fable 5 is portrayed as stronger on atmosphere, detail, and visual appeal, winning decisively on a Skyrim-style Frostfell game and looking smoother and more interesting in a Remotion video-creation build. Overall results are very close, and the narrator says model choice depends on tooling, preferring Fable 5 largely due to Claude Desktop, while recommending running both models together inside an Agent OS workflow.

00:00 Sol vs Fable Setup
00:52 Dragon Game Showdown
01:36 Dragon Realm and Doom
02:22 3D Racer Verdict
03:09 Crypt Dungeon Test
03:34 Skyrim Style Frostfell
04:51 Coding Tool Pick
05:27 Agent OS Workflow
06:16 Costs and OAuth
06:57 Goldie Bench Builds
07:14 Flight Sim and Neon City
08:08 Remotion Video Test
08:47 UX and Codex Sites
09:17 Run Both in Agent OS
10:10 Join the Community
10:54 Final Thanks

The video explains an “agent operating system” as a single place where multiple AI agents work together to automate tasks and move from idea to implementation quickly, using tools like Paperclip group chat, a one-click pipeline, and integrated models such as Claude, Fable 5, and GPT 5.6 (via API or OAuth). It highlights Hermes and related modules (Jarvis-style voice agent, Mission Control, Oracle, Astros), plus built-in workflows for lead outreach, SEO keyword discovery and publishing, video and music creation, and a shared memory system that keeps all agents in the same context. The episode is framed around recent AI Profit Boardroom community questions and positions the Agent OS as a customizable, time-saving system for building and automating with AI agents.

00:00 Agent OS Overview
00:17 Core Features Tour
01:27 Codex and Model Switching
02:09 Multi Device Setup VPS
03:35 AI Avatar Creation Tips
04:09 Hermes Mission Control
05:03 Cheaper Writing Models
05:50 AI Marketing Agency Playbook
07:17 Mobile Access On The Go
07:53 Community Partnerships
08:05 Obsidian Memory Vault
09:47 Wrap Up and Next Steps

The video introduces GPT 5.6 integrated into an Agent Operating System that uses OAuth with an existing ChatGPT subscription, requiring no API key, and provides a single dashboard to switch between models and tools like GPT 5.6, Codex, Claude, Hermes, Groq, and free options such as OmniRoot/Hy3 when tokens run out. It highlights five benefits of the “GPT 5.6 command engine”: one login, tier switching (Sol, Terra, Luna), goal mode for autonomous background task loops in a sandbox with optional auto-approve, a workspace where all builds are saved and previewable, and a unified command center with memory so nothing is lost. The script argues this setup is more organized and productive than juggling separate apps or tabs and can be customized for different business workflows.

00:00 GPT 5.6 Overview
00:48 Unified Agent OS
01:26 Why Tabs Fail
02:22 Five Key Benefits
03:13 Old Way vs New
04:24 OAuth Setup Guide
04:49 Model Tier Choices
05:34 Goal Mode Automation
06:27 Build Quality Tests
07:06 Custom Workflows
08:08 Speed and Tokens
08:59 Recap and Proof
10:07 Join the Community
11:31 Final Thanks

The script reviews OpenAI’s GPT-5.6 rollout and its three model sizes—Sol (flagship for hard multi-step agentic work), Terra (everyday production), and Luna (fast, cheap, high-volume)—and shows results from extensive testing on Goldy Bench, including games, a full video animation, landing pages, and dashboards. It notes shared specs (1M context window, 128K max output, Feb 2026 knowledge cutoff) and compares performance, speed, and cost across the three on the same 3D space game task, where Sol is slower and more expensive but produces larger, higher-quality output. The presenter compares Sol against “Fable 5,” recommends switching models by task tier to control cost, suggests using Codex/Agent OS and Hermes over ChatGPT for better output, and promotes the AI Profit Boardroom community and tools.

00:00 GPT 5.6 Overview
00:32 Sol Creations Showcase
02:20 Video Generation Demo
03:07 Specs Pricing Access
04:53 Benchmarks Reality Check
06:03 Speed Cost Comparison
07:17 Web Pages Dashboards
08:41 Model Routing Strategy
09:57 Agent OS Workflow
11:45 Fable 5 Versus Sol
13:12 AI Profit Boardroom Pitch
14:52 Wrap Up Thanks

The video introduces Grok 4.5, newly released, and demonstrates a free agentic operating system built around it using X/Twitter, showing step-by-step how it plugs into a CLI and saves everything in a workspace. The creator highlights Grok 4.5’s strengths for coding, agentic tasks, and knowledge work, and shows it building a landing page plus visual projects like games and automated Remotion-based videos. Grok 4.5 is integrated into Hermes Agent with OAuth login, a memory system synced via Obsidian, and a “Hermes Oracle” that pulls real-time trending news from Twitter, categorizes it by virality, and enables one-click content creation and publishing (including WordPress). The system supports mixing models (Claude, others) and offers access via the AI Profit Boardroom with tutorials, support, and coaching.

00:00 Grok 4.5 Overview
01:03 CLI Build Demo
01:42 Games and Video Creation
02:21 Hermes Agent Integration
04:18 Video Agent Workflow
05:24 Image and Video Prompts
06:28 Twitter News Oracle
07:37 Why Systems Beat Models
08:14 Landing Page Results
08:29 Swap Models and Free APIs
09:02 Get the Agent OS
09:41 Community and Coaching
10:33 Wrap Up

The video compares Grok 4.5 and Claude Opus 4.8 side by side using Goldiebench, emphasizing real-world performance over benchmarks across multiple game and creative tasks. Grok 4.5 generally produces more playable, smooth 3D games and wins many rounds where Opus 4.8 fails to generate or is too buggy, though Opus performs better in some cases like a driving game, an edited Remotion video, and a flight simulator, and also wins a few specific games. The creator notes Grok can be slow and less agentic in chat and wouldn’t use it for agent workflows unless generating images/videos, while preferring Opus 4.8 for building systems, workflows, and an Agent OS. The video ends by promoting the AI Profit Boardroom for access to Agent OS, agents, tutorials, and community support.

00:00 Grok vs Opus Setup
00:38 Why Real Tests Matter
00:57 Crypt Game Showdown
01:34 Dragon Realm Faceoff
02:15 Racing and Skyrim Tests
03:41 Twilight and VoxelCraft
04:43 Runner and Driving Results
05:26 Video Edit and Flight Sims
06:36 Agentic Use in Hermes
07:04 More Games and Scoreboard
09:24 Final Verdict and Picks
10:39 Boardroom Offer and Outro

The script reviews the newly released Grok 4.5, noting it’s free to try and showing results from eight game builds tracked on Goldy Bench, with seven good outputs and one poor one. The presenter demos multiple genres (open-world RPGs, racing, voxel/Minecraft-like) and comments on graphics, physics, controls, occasional bugs, and smoothness, plus the benefit of access via an existing X/Twitter subscription. They compare Grok 4.5 against models like Fable 5 and Opus 4.8, finding Fable 5 often higher quality while Grok 4.5 sometimes matches or beats Opus 4.8, including cases where Opus fails to generate. The script also shows Grok 4.5 integrated into Hermes/Agent OS, CLI, and studio workflows for images, video, and news-driven automations, and mentions token efficiency, real-time Twitter news access, and community resources.

00:00 Grok 4.5 Launch
00:08 Eight Game Builds
00:32 RPG and Racing Demos
01:38 Voxelcraft and Fail
02:29 Seven of Eight Verdict
02:48 Model Comparisons
06:12 Benchmark Results
07:06 Hermes Integration
08:17 Studio Image Video
09:47 Custom Workflow Power
10:06 Hermes Oracle News
12:37 Speed Tokens Pricing
14:59 Memory Galaxy Loop
15:31 Community and Wrap

The script reviews using Grok 4.5 inside Hermes Agent, noting free access via Twitter subscription and testing across about 45 builds. Grok 4.5 is described as close to Claude Opus 4.8 and better than GLM 5.2 for competing with Opus, but it’s criticized for slow tool-calling in chat (often 10–120 seconds), making it less suitable as a direct agent profile. The recommended approach is using Grok 4.5 in custom, SaaS-style workflows inside an Agent OS: one-click image/video/voice generation, B-roll creation, and real-time research via Twitter’s API, while using faster models like GLM 5.2 for chat. The script also showcases Hermes Oracle for daily trending-news research and one-click SEO blog publishing, plus automatic memory updates to an Obsidian knowledge graph, and ends promoting the AI Profit Boardroom community and Agent OS resources.

00:00 Grok 4.5 Overview
00:47 Tool Calling Speed Test
01:46 Best Workflow Setup
02:33 One Click Image Studio
03:17 B Roll Video Workflow
05:29 OAuth Cost Advantage
06:32 Hermes Oracle News Engine
07:35 Auto Updating Memory
08:52 Showcase Builds Demos
09:40 Community And Access
10:58 Final Thanks

The episode answers member questions on using an Agent Operating System that combines a memory system with recent model and tool updates, including a Grok 4.5 CLI, integrations like free Claude code via HY3/OmniRoot, and Hermes components such as Oracle, Astros, and a voice-powered agent. It explains that Rough Flow is now largely unnecessary due to Claude UltraCode and Hermes’ built-in sub-agent teams with Kanban visualization. The host shares practical setup guidance for OpenSEO via GitHub and Claude, compares hosting Hermes via Abacus versus self-hosting or Hermes Cloud, recommends implementing AI within an existing business, advises keeping a single Obsidian vault to reduce complexity and tokens, and outlines a “second brain” workflow using Omi, Obsidian, and Memory Galaxy. It also discusses in-demand automations, token reduction tactics, and when Paperclip may be unnecessary.

00:00 Agent OS Overview
01:04 Rough Flow vs Hermes
02:06 OpenSEO Setup Guide
02:47 Hosting Hermes Options
03:25 AI in Your Business
04:13 Obsidian Vault Strategy
04:52 n8n Carousel Workflows
05:16 Install Agent OS Hermes
06:09 GoHighLevel Contacts Import
06:20 Safe Agent OS Setup
07:39 Second Brain Memory Galaxy
09:15 Selling AI Automations
10:59 New Model Updates
11:22 Cut Hermes Token Costs
12:01 Paperclip with Hermes
13:27 Wrap Up and Community

The script explains how to prepare for OpenAI’s GPT 5.6 public launch on Thursday, which includes three models—Sol (flagship), Terra (lower cost), and Luna (fastest and cheapest)—with preview access expanding globally. It outlines four prep moves: log every daily task, time-audit tasks by duration and frequency, minimize tokens before scaling (using free/local models and token “playbooks” like Caveman, Ponytail, RTK, Headroom, and techniques such as shorter paid replies and matching reasoning effort to the task), and build an agentic operating system around the audit so GPT 5.6 can plug into existing automations immediately. The script argues the system around the model matters more than the model itself and promotes the AI Profit Bot as a ready-made Agent OS with agents, memory, free APIs, tutorials, community support, and coaching calls.

00:00 GPT 5.6 Launch Hype
00:39 Three Model Breakdown
01:12 Four Prep Moves
03:21 Audit Your Week
04:15 Token Saving Playbooks
04:43 Pick the Right Model
05:16 Cut Costs and Effort
07:21 Systems Beat Models
08:27 AI Profit Bot Offer
09:45 Wrap Up

The script explains how to prepare for OpenAI’s GPT 5.6 public launch on Thursday, which includes three models—Sol (flagship), Terra (lower cost), and Luna (fastest and cheapest)—with preview access expanding globally. It outlines four prep moves: log every daily task, time-audit tasks by duration and frequency, minimize tokens before scaling (using free/local models and token “playbooks” like Caveman, Ponytail, RTK, Headroom, and techniques such as shorter paid replies and matching reasoning effort to the task), and build an agentic operating system around the audit so GPT 5.6 can plug into existing automations immediately. The script argues the system around the model matters more than the model itself and promotes the AI Profit Bot as a ready-made Agent OS with agents, memory, free APIs, tutorials, community support, and coaching calls.

00:00 GPT 5.6 Launch Hype
00:39 Three Model Breakdown
01:12 Four Prep Moves
03:21 Audit Your Week
04:15 Token Saving Playbooks
04:43 Pick the Right Model
05:16 Cut Costs and Effort
07:21 Systems Beat Models
08:27 AI Profit Bot Offer
09:45 Wrap Up

The script explains five autonomous “Claude agent loops” built inside an Agent OS that reduce manual prompting by using a repeatable cycle where an agent acts, an independent judge checks the work against a defined goal, and the loop repeats until it passes, optionally with a human approval step. The five loops include: a loop-engineering builder/judge setup to build deliverables to a spec while optimizing cost with different models; a Kanban workflow with planner, builder, and reviewer roles that moves tasks through triage, running, blocked, and done; a pipeline that takes ideas from inbox to shipped with one-time plan approval and saved outputs in a vault; an “Oracle” timed news loop that finds and ranks AI news and produces content hooks while logging to Obsidian memory; and an SEO pipeline that generates keywords and one-click SEO articles. A shared memory system updates automatically to compound performance, and the video promotes access via the AI Profit Boardroom with coaching and community support.

00:00 Autonomous Agent Loops
02:26 Old Way vs New Way
03:23 Loop Engineering Builder
04:22 Kanban Multi Agent Board
05:25 Idea to Shipped Pipeline
07:20 Oracle News Watcher
08:42 SEO Content Pipeline
09:10 Shared Memory System
09:55 Wrap Up and Access

The episode demonstrates a system that adds a full AI video editor inside Agent OS using a free open-source GitHub skill, allowing users to upload footage, prompt an agent, and generate edited videos with before-and-after results. The workflow can be taught to models like Hermes or Claude and is positioned as automating tasks typically handled by a team of video editors, functioning like a SaaS-style “video agent” that saves projects in a workspace for easy organization. It emphasizes writing strong briefs to improve edits, notes that rendering can take time so users should leave it running in the background, and mentions optional integration with ElevenLabs. The system can also create full avatar videos with B-roll, voiceover, and research in one prompt, leveraging memory and context for personalization, and is available as a full setup through the AI Profit Boardroom.

00:00 Video Editor Inside Agent OS
00:39 Open Source Video Skills
01:27 Video Agent Workflow
02:15 Install in Five Minutes
02:36 Writing Better Briefs
02:59 Processing Time and Tips
03:30 AI Avatar Video Examples
03:42 Full Video Team System
04:11 Workspace Memory and Context
04:39 Join AI Profit Boardroom
04:46 Community Training and Support
05:08 Coaching Calls and Networking
05:22 Final Wrap Up

The video shows how to use Codex for free by plugging Tencent’s new open-source HY3 model into Codex via a free OpenRouter API (available for about two weeks, until around July 21). The creator demonstrates a custom setup with goal mode and a workspace where you can preview and manage past builds, plus a separate “HY3 Coder” that runs directly on HY3 without Codex. They explain the workflow: Codex acts as the hands (writes files, runs commands) while HY3 is the brain (agentic coding and tool calling), and you can swap the model in the config or have Claude/Claude Desktop set it up. If rate-limited, you can switch to OmniRoot, another free API, and also use HY3 inside Hermes as a free agent profile.

00:00 Free Codex With HY3
00:35 Workspace Demo
01:08 Setup Via OpenRouter
02:10 HY3 Coder Alternative
02:44 Hands And Brain Explained
03:36 Hermes And OmniRoot
04:29 Live Task Switching
05:23 HY3 Performance Benchmarks
06:11 Token Saving Workflow
07:16 Get The Full Agent OS
08:22 Final Thanks

The video covers Tencent Hunyuan HY3, a new free, open-source 295B-parameter mixture-of-experts model from China, available on Hugging Face and reportedly outperforming DeepSeek and Qwen on some benchmarks. The host explains two free ways to run it (including a free OpenRouter API trial until July 21) and shows using it inside an agent operating system to generate code like a landing page in real time, as well as integrating it into Hermes tools via a portal offering a two-week free window. They share honest comparisons from their own tests versus GLM 5.2 and models like Opus 4.8 and “Fable 5,” concluding HY3 can be useful for agentic and auxiliary sub-agent tasks but doesn’t replace top models, though it impressed by generating a fully animated video via Remotion. The episode ends with a pitch for the AI Profit Ballroom community, guides, and the agent OS.

00:00 HY3 Model Overview
00:19 Benchmarks and Hype
00:42 Run It Free via OpenRouter
01:42 Hermes Integration Options
02:16 Honest Model Comparisons
02:49 Real Project Demos
05:00 Best Use Cases and Workflows
06:05 Agentic Focus and Bench Claims
06:31 Video Creation Test
07:28 Wrap Up and Next Drops
07:58 Community and Guides Pitch
08:54 Final Thanks

The script explains the creator’s Claude Agent Operating System, a unified setup where multiple agents (e.g., Hermes, OpenClaude, Claude, Antigravity, Codex) share a common memory that auto-logs updates so agents learn the user’s voice, brand, goals, projects, and clients. It showcases recent additions like Hermes Apollo (a voice-activated agent), Hermes Oracle (pulls trending news and can publish content to a website), and Hermes Astros (keyword/topic monitoring that generates daily content ideas and routes prompts to agents). The system includes a command deck for mission control, model layers with premium and free options, an agent fleet and Kanban workflow from plan to ship, and token-minimization tools (RTK, Caveman, Ponytail). It also features a content factory, studios for design/games/AI movies, and an Obsidian-based “memory galaxy,” and is offered via the AI Profit Boardroom with tutorials, community support, and coaching.

00:00 Agent OS Overview
01:01 Voice Agent Hermes Apollo
01:42 News Oracle Automation
02:38 Keyword Ideas Engine
03:23 Kanban Agent Teams
03:50 New Integrations Added
04:23 Access and Installation
04:46 System Layers Explained
04:54 Command Deck Mission Control
05:21 Models and Agent Fleet
06:19 Ideas Pipeline Workflow
06:44 Token Diet Strategies
07:37 Content Factory Automation
08:26 NotebookLM Content Library
09:15 Studios Design Games Movies
09:56 Memory Galaxy in Obsidian
10:48 Community Wins and Proof
12:02 Why Systems Beat Models
13:08 Boardroom Support and Coaching
13:51 Classroom Training and Wrap Up

The script compares GLM 5.2 (from ZAI) versus Tencent Hunyuan HY3 (HY3) side by side using multiple generated projects, including game-like demos (Dragon Realm, flight sim, driving game, parachute game) and a Remotion cinematic video. HY3 is described as buggy, laggy, and often lacking detail, requiring hours of back-and-forth to get usable outputs, while GLM 5.2 generally produces smoother, more playable results on the first try. HY3 wins one round where GLM 5.2 fails, but GLM 5.2 wins 6 of 7 tests overall and also has a larger context window (1M tokens vs 256K). HY3’s advantages are free API access (via OpenRouter until July 21) and availability via Hermes Agent, while GLM 5.2 is preferred overall.

00:00 GLM vs HY3 Intro
00:26 Free Access Setup
00:52 Test 1 UI Smoothness
01:49 Dragon Realm RPG Test
02:50 Flight Simulator Showdown
03:25 Neon City Driving Test
04:12 When GLM Fails
04:41 Parachute Game Comparison
05:05 Remotion Video Ad Test
05:45 Scorecard and Context
06:13 HY3 Strengths and Tips
07:00 Final Verdict and Outro

The script demonstrates how to run Hermes Agent for free using the newly released open-source OmniRoute/Omniroot gateway, which routes requests across 231–237 providers with millisecond auto-fallback when a free API is rate-limited. It explains that OmniRoute compresses tokens using RTK (removes repetition/duplication) and Caveman (short, blunt outputs) to reduce rate limits, and offers one endpoint that Hermes can use without “knowing the difference.” The creator shows testing Hermes with an OmniRoute profile, building a landing page locally, and recommends using separate Hermes agent profiles per API/model to compare side by side. Setup is described as non-technical via simple terminal commands and GitHub instructions, with examples of free-forever providers and optional OpenRouter integration. The video also promotes the AI Profit Boardroom and its Hermes Agent OS workflows, memory system, community support, and coaching calls.

00:00 Free Hermes Setup
01:01 OmniRoute Benefits
01:43 Hermes Profiles Demo
02:57 How Routing Works
03:55 Workflows and Memory
04:33 Dashboard Setup Steps
05:51 Free Providers List
06:29 Token Compression Explained
07:01 Wrap Up and Results
07:23 Agent OS Offer
08:24 Community and Coaching
08:50 Final Thanks

00:00 HY3 Free Release
00:35 Quick Demo Setup
00:53 Install and Connect
01:16 Free API Options
01:47 Landing Page Test
02:08 Tool Use and Learn
03:03 Workflow Summary
03:31 Benchmarks and Limits
04:18 Token Saving Strategy
05:11 Agent OS Pitch
06:04 Community and Tutorials
07:02 Final Call to Action

The video tests Tencent HY3, a new open-source model available for free via OpenRouter until July 21, and shows it integrated into Agent OS plus availability in Hermes Agent, News Research, and Kilo Code. The presenter demos HY3 building a landing page and several game-like creations, noting outputs often look basic, glitchy, and less detailed or playable than results from models like Fusion, GLM 5.2, Claude, Fable 5, and Opus 4.8, despite HY3’s benchmark scores appearing strong on some agentic evaluations. HY3 performs better for agent workflows such as browsing, learning skills, running tools, and generating Remotion-based animated videos, and the HY3 Coder setup is highlighted as a useful free workflow. Technical notes include HY3 being a 295B parameter mixture-of-experts model designed for agent capabilities, with local-running details still emerging.

00:00 Meet Tencent HY3
00:22 Free API Access
00:56 Landing Page Demo
01:13 GoldieBench Game Tests
01:47 More Demos And Issues
03:26 Benchmarks Reality Check
03:56 Comparing To Top Models
04:50 Best Use Agent Tasks
05:29 HY3 Coder Workflow
05:47 Remotion Video Creation
06:44 Specs And Local Running
07:31 Final Verdict And Next Tests
07:54 Agent OS And Community Plug
09:01 Wrap Up

The video tours a custom “Hermes Agent OS” the creator built to make Hermes far more powerful than using it only in the terminal or desktop app. It showcases Hermes Astros, a workflow that monitors competitors and keywords via an API and turns trending topics into content ideas that can be sent in one click to a video agent, NotebookLM, or SEO content. It also demonstrates a Hermes-based outreach tool for on-demand lead generation, validation, enrichment, and AI-built email campaigns. Hermes Apollo is shown as a fast, voice-activated assistant that can open apps, use the browser, build mini apps/websites, run in agent mode, use a wake word, and deliver daily/weekly briefings. Hermes Oracle pulls trending industry news, ranks topics, and can generate and publish first-person articles with formatting and CTAs. The system also includes a studio for media generation, a “mixture of agents” setup combining models, and broader orchestration tools, with access offered via the AI Profit Boardroom.

00:00 Hermes Agent Overview
00:24 Astros Trend Monitor
01:44 Outreach Lead Engine
02:49 Apollo Voice Assistant
04:37 Oracle News to Content
06:36 Studio and Agent Mix
08:15 Why Custom OS Wins
08:54 Orchestration and Memory
09:59 Get the Agent OS
10:55 Wrap Up

This episode answers recent AI Profit Boardroom questions about configuring and using the Agent Operating System, including how to set up individual CLIs from the install folder (and hide unused ones), and how to organize client work using an Obsidian memory system with separate project folders or custom workflows. It covers why adding too many skills/rules can slow agents down and shows using forward-slash goal mode in tools like Claude Code and Codex, where a judge model checks progress each loop. The script discusses whether to run local LLMs and Agent OS on one machine or separate machines, recommends keeping them local on one device for simplicity and security, and shares community resources for building a connected Obsidian knowledge graph. It also explains setting up iMessage in Hermes via Channels/BlueBubbles and installing FreeClaude Code and OmniRoot for free coding across many APIs.

00:00 Agent OS Q&A Intro
00:18 Configuring CLIs Fast
01:10 Client Projects Setup
02:01 Autonomy Without Slowdowns
03:09 One Machine Or Two
03:59 Connecting CLIs And Agents
04:36 Obsidian Knowledge Graph
06:04 Hermes iMessage Setup
06:51 Free Claude Code Tour
07:41 OmniRoot Free Coding
08:33 Wrap Up And Community

The script explains testing RTK, an open-source filter that sits between an AI agent and the shell to remove padding and duplicate output from common commands, reducing token usage while keeping results the same. The presenter reports 82.9% average token savings in their tests, cites RTK’s README claim of 60–90% savings, and notes minimal added latency (~14 ms). An example shows swapping git diff to an RTK version that strips repeated headers and unchanged lines, cutting output from 373,000 to 29,000 characters (92% less) while saving full unshortened output to a file if needed. RTK is described as easy to install on Linux/macOS, small (6.6 MB), and capable of shortening outputs for about 100 everyday commands. The script also covers stacking RTK with other open-source tools—Ponytail, Headroom, and Caveman—to further minimize tokens, and promotes the Agent OS setup and AI Profit Bot community.

00:00 RTK Token Savings Intro
00:55 How RTK Works
01:47 Real Test Results
02:28 Step by Step Example
04:42 Testing Method
05:07 Install and Command Support
05:38 Does It Lose Info
05:51 Token Minimizer Stack
06:36 Caveman Mode Demo
07:39 Fable 5 Efficiency Wins
08:23 Agent OS and Community
09:25 Wrap Up

The video demonstrates a “free Codex engine” setup that uses OpenAI’s Codex CLI routed through OmniRoute/OmniRoot to access 93 providers (with auto fallback when a free API runs out or rate limits), allowing Codex to plan, write files, and run commands as usual without a Codex subscription. The creator shows building a sleek SEO agency landing page and a working to‑do list app in about 60–90 seconds, with previews and a workspace to revisit past builds inside an Agent OS environment that can also connect outputs to tools like Claude Code and other agents. The script explains that this approach is cloud-based, lightweight on a basic laptop, and quicker than local models, and claims setup is only a couple of commands. It ends by promoting the AI Profit Ballroom community and training for installing the Agent OS zip and learning OmniRoot/Codex workflows.

00:00 Free Codex Overview
00:42 Agent OS Demo
01:45 Why OmniRoot Wins
03:28 Local Models Comparison
04:10 How Routing Works
05:09 To Do App Test
05:59 Old vs New Setup
06:53 Three Command Install
07:52 FAQs and Token Tips
08:55 Goal Mode Idea
09:23 Workspace Features
09:41 Get the Full Setup
10:12 Community and Training
10:53 Final Call to Action

The episode answers recent questions about building and using an agent operating system, emphasizing that once set up it can transform daily work by automating recurring tasks and integrating useful open-source tools like AI film generation and SEO systems. For beginners, the advice is to simplify by only adding the tools you actually use (e.g., just Claude), automate what currently consumes the most time, and personalize the UI and features to increase adoption. For installation on a VPS, viewers are told they can ask their agent to run through the guide’s modules and should avoid installing unnecessary components. The script also discusses staying productive despite “shiny object” distractions, using a video agent and SEO tools for content creation while posting manually to avoid platform penalties, leveraging Hermes Oracle for trending topics and WordPress publishing, teaching workflows via Claude’s browser extension, debugging Paperclip with Claude, and highlights shared memory/knowledge graph, Hermes Astros keyword monitoring, and an AI “SEO office” of agents available through the AI Profit Boardroom community and coaching.

00:00 Agent OS Overview
01:13 Start Simple First
02:32 Install Checklist Tips
03:24 Boredom And Focus
04:09 Content Automation Stack
06:06 Teach Agents Workflows
06:36 Fixing Paperclip Bugs
07:20 Hermes Agents Demo
09:13 Get The Full Setup
09:52 Final Thanks

This episode demos OmniRoute, a free open-source AI gateway that lets coding tools like Claude Code and Cursor route through a single local endpoint to 237 providers, including 90 that are free forever, with automatic fallback when limits are hit. The presenter shows OmniRoute generating projects (an animated star field, a to-do list app, a virtual keyboard, and a landing page) with code and live previews, and highlights token-minimization that can cut usage by 15–95%. Setup is shown as quick via simple terminal commands, runs across 42 languages, and supports variants like auto/fast and auto/coding plus 17 routing strategies including Fusion. The video also explains how OmniRoute is integrated into an Agent OS “free coding engine” with saving/visualizing builds, and closes by promoting the AI Profit Boardroom community, trainings, and coaching calls.

00:00 OmniRoute Overview
01:10 Setup in Agent OS
01:40 Quick Install Steps
01:59 Building a To-Do App
02:17 Why Free Tiers Fail
03:09 Key Benefits Recap
03:22 Five Layer Routing
04:01 Speed Demo Build
04:56 Provider Disappear Fallback
05:35 Models and Routing Modes
06:07 Fusion Strategy Explained
06:33 Full Agent OS Pitch
07:01 Community and Training
07:44 How to Join and Outro

The video reviews Agent A1, a newly released free Chinese local model designed for agentic tasks, showing it running privately and offline inside an Agent Factory workflow alongside free Claude code. The host demos outputs like a landing page, a neon snake game, a “Dragon Realm” open-world game, a keyboard mini app, and a 3D solar system visualization, noting fast replies and that builds save to a workspace with optional voice control. Agent A1 is a 35B mixture-of-experts model (3B active) with a 256k context window, created by Intern Science in Shanghai and optimized for long-horizon search, engineering, scientific research, instruction following, and tool calling. It ranks #2 on the creator’s local leaderboard, outperforming Gemma 4 but trailing Qwable, and is reported around 95 tokens/sec locally on a Mac Studio M4. The video explains strengths/weaknesses, shows running via Ollama and configuring Hermes profiles, and promotes an Agent OS bundle plus OmniRoute routing across 90 free models.

00:00 Meet Agent A1
00:31 Live Local Build Demo
00:53 Leaderboard And Examples
01:36 What A1 Actually Is
02:29 Benchmarks Breakdown
03:08 More Builds In Action
03:45 Specs Speed And Context
04:40 How To Run Locally
04:43 Strengths And Weaknesses
05:41 Hermes Setup Steps
06:13 Agent OS And OmniRoute
06:51 Get The Full Setup
07:34 Wrap Up

The script explains a free, open-source “Caveman” skill that reduces output tokens for Claude Code and other agents (Codex, Gemini, Cursor, Windsurf, Cline, Copilot, and more) by making responses short, blunt, and direct while keeping code, commands, file paths, and error messages unchanged. Installed via a one-line command, it drops a rules text file into each agent’s skills/plugin folder so the agent reads it at the start of every chat. Tests on Fable 5 using five real prompts showed about 69% fewer output tokens (e.g., 1,349 down to 324) and about 37% lower total cost while keeping answers correct; it can be toggled with /caveman and set to light/full/ultra, plus optional tools like Commit/Review/Compress. The video also promotes using Caveman across an agent operating system for compounding savings and mentions the AI Profit Ballroom community and training.

00:00 Cut Tokens With Caveman
01:16 Why Token Costs Hurt
01:58 How Caveman Works
02:21 Rules And Safety
03:06 Does Shorter Mean Worse
04:00 Install Across Agents
05:03 Real Token Savings Tests
07:29 Old Replies Vs New
08:49 Modes And Commands
09:47 Agent OS And Extras
10:30 Join The Community
11:29 Wrap Up

The script demonstrates Hermes Apollo, a voice-activated assistant inside Agent OS that responds quickly via ChatGPT real-time and can run agentic tasks through a backend agent using GLM 5.2. The presenter shows Apollo telling a joke, teaching a Chinese word (“xie xie”), building a snake game and a 24-hour countdown timer with previews, and opening a website by voice. Apollo supports wake-word hands-free use, wall mode, chat and build history, weekly/monthly briefings, memory recall via an Obsidian “memory galaxy,” and can switch to agent mode to control the browser. The broader system includes Hermes Oracle for trending Twitter news and Hermes Astros for competitor monitoring and content idea generation. The presenter emphasizes speed, usefulness beyond Siri, local/private operation options, and invites viewers to get the full setup in the AI Profit Boardroom with tutorials, guides, coaching calls, and community support.

00:00 Meet Apollo Demo
00:28 Voice Commands Build Apps
01:00 Jarvis Style Overview
01:33 Briefings And History
02:17 How It Works Backend
02:51 Modes And Wake Word
03:18 Agent Mode Browser Control
03:51 Prompts And Use Cases
04:15 Memory With Obsidian
04:38 Oracle And Astros Tools
05:43 Not Just A Toy
06:03 Privacy And Local Models
06:18 Setup Proof And Testimonials
06:34 Get Agent OS Access
07:02 Community Coaching And Wrap

The video explains how to reduce token usage when working with Fable 5, noting it can consume tokens quickly, only allows 50% of a subscription, and will move off-subscription by July 7. It shares 10 practical tactics, including routing tasks to cheaper models (e.g., using Fable 5 only for the hardest work, and Opus/Haiku/GLM 5.2 for general coding), using Fable 5 for planning while a cheaper model implements, lowering the effort level, and installing open-source tools like Headroom (60–95% token savings) and Ponytail (54% less code on average, up to 94%). Additional tips include trimming Claude MD/rules, turning off web search by default, using /compact plus a brief handoff to preserve context, and doing a periodic /clear with a short note to avoid long-session token bloat. The episode also highlights the creator’s Agent Operating System and AI Profit Boardroom community for token-efficiency playbooks, training, and coaching.

00:00 Why Tokens Matter
01:01 Route Tasks by Model
01:45 Plan Then Build Cheap
02:58 Adjust Effort Settings
03:47 Headroom Token Shrinker
04:35 Ponytail Lazy Dev
05:39 Trim Rules and Search
06:48 Manual Compact Workflow
08:08 Two Hour Clear Reset
08:58 Wrap Up and Results
09:22 Agent OS Pitch
10:26 Community and Outro

The speaker demos a custom “Claude agent operating system” built over two months to unify tools like Fable 5 and Hermes into a single, customizable mission-control UI with shared memory and organized outputs. They show how features evolved from simple chat into multiple model profiles, Hermes Mixture of Agents, a wake-word voice agent that can control the browser and local actions, and Hermes Oracle for pulling and linking breaking news and drafting social content. The system includes agent orchestration via group chat, pipelines, and a Paperclip-style org chart with departments, an inbox, dashboards, and progress history, plus Kanban workflows and integrations like NotebookLM video generation. It’s designed to swap models/CLIs quickly, run scheduled “loop” automations (outreach, news refresh), and maintain a growing memory/notes system. The OS and tutorials are offered through the AI Profit Boardroom.

00:00 Agent OS Overview
00:21 Why Mission Control
01:12 Profiles and Mixture
02:47 Keep It Simple
03:11 Voice Agent Demo
04:14 Oracle News and Content
04:29 Building With Claude
05:18 Agent Mastermind Chat
06:05 Paperclip Agent Org
08:07 Swap Models Fast
09:16 Kanban and Video Tools
10:12 Loop Engineering Automation
11:24 Memory Feedback System
12:12 Community and Access
13:14 Courses Coaching Wrap
14:00 Final Thanks

The video compares Anthropic’s Fable 5, Hermes Mixture of Agents, and Open Router’s Fusion by testing 42 builds and showing side-by-side results across multiple projects, including a racer game, Nordic crypt game, orbit simulation, landing page, Dragon Realm, a WebOS, VoxelCraft, Frostfell, and a Doom-style game. Fable 5 wins the racer game, crypt game, and WebOS for usability and stability, but loses the orbit test to Hermes and the landing page test where Hermes produces the preferred design and Fusion beats Fable 5. Fusion wins VoxelCraft, while Frostfell is considered a tie, and Fable 5’s Doom build fails basic movement, giving the edge to Hermes/Fusion. The creator suggests mixing Mixture of Agents with Fable 5, recommends Hermes as a backup when Fable 5 tokens run out, notes Fable 5 subscription availability until July 7, and promotes tools, guides, and community access via AI Profit Boardroom and Goldie Bench.

00:00 Showdown Setup
00:18 How Agents Combine Models
01:04 First Game Comparisons
01:42 Nordic Crypt Results
02:48 Orbit Simulation Test
03:41 Landing Page UI Battle
04:42 Why Mixture Works
05:31 Dragon Realm Tie
06:27 WebOS Usability Test
08:30 VoxelCraft and Frostfell
09:53 Doom Clone and Scorecard
10:51 Final Takeaways
11:59 Community and Resources

The script reviews Claude’s Fable 5 after its return, showing results from 42 builds and how to access it by selecting Fable 5 in Claude’s model list. It notes Fable 5 is included in a subscription only until July 7, after which it may require extra paid credits, and it sometimes auto-switches to Opus 4.8—especially on science-related prompts. The creator demonstrates strong coding and project-building speed (including a web “operating system”) but highlights weaknesses such as poor web UI output, buggy first builds, and inconsistent instruction-following compared with a custom-trained Opus 4.8. Side-by-side tests show panel approaches (Hermes Mixture of Agents, Fusion, Sakana Fugu) can outperform Fable 5 on some games and simulations, and the video promotes access to these systems and tutorials via the AI Profit Boardroom community.

00:00 Fable 5 Returns
00:32 Access and Pricing Window
01:15 Showcase Web OS Build
01:38 Panels vs One Shot Tools
02:08 UI Weaknesses and Bugs
03:15 Tutorial Skill Compliance
04:33 Backlash and Model Switching
05:31 Side by Side Build Tests
08:07 Alternatives and Benchmarks
08:41 Community and Training Offer
09:52 Wrap Up and Links

This episode answers recent community questions about the Agent Operating System (Agent OS), which centralizes and orchestrates multiple AI agents in one place. It covers recent updates including a Hermes lead generation tool, Mixture of Agents testing, an auto-updating memory system using Obsidian, a new GLM Code section to use GLM 5.2 with agent harnesses like Claude Code, NotebookLM short video generation and research import, an expanded SEO content pipeline with OpenSEO, and plans to add Fable 5 as a default CLI when restored. The host advises staying focused with a “Focus Protocol,” recommends Agent OS for managing multiple tools, shares guidance on Docker and GitHub/data concerns, compares models (preferring Opus 4.8, GLM 5.2 over Sonnet 5), suggests SEO stacks for local businesses, and highlights community wins, customization examples, and how to join AI Profit Bomb for training, support, and the full Agent OS.

00:00 Agent OS Updates
01:51 Focus Protocol
03:21 Why Use Agent OS
04:49 Docker and GitHub
06:47 Model News and Picks
07:52 SEO Side Gig Stack
08:58 Cheaper Models Setup
09:32 Community Wins Workflows
11:45 Free Models OwlAlpha
12:33 Themes and Customization
13:41 Best Memory System
14:48 Kanban Orchestration
15:44 Ollama vs Hermes
16:27 More Memory Advice
17:22 Custom Desks Example
18:30 Join the Community

The episode covers the official release of LongCat 2.0, an open-source Chinese agentic model revealed as the model behind the AoAlpha free API, with features like Sparse Attention, Zero Compute Experts, and MIPD. The host reviews benchmark claims (including Terminal Bench 2.1 and SWE-Bench Pro comparisons versus GPT-5.5 and Opus 4.8) and shares hands-on tests building game demos such as Dragon Realm, a Skyrim-style open world, and VoxelCraft, noting mixed results and frequent bugs. Access issues are mentioned, including difficulty using the API without a Chinese setup, so the model is tested via the website chat. A key point is that LongCat was trained on China’s Meituan chips without NVIDIA. Overall, GLM 5.2 is judged stronger in side-by-side game benchmarks, and the host promotes the AI Profit Boardroom and Agent OS setup.

00:00 LongCat 2.0 Launch
00:36 Benchmarks and API Hurdles
01:38 Game Demos Dragon Realm
02:23 Goldy Bench Verdict
02:43 Trained Without NVIDIA
03:32 How to Use It
03:51 Eval Results vs GPT
04:17 GLM 5.2 Showdown
06:13 Final Take and Recommendation
06:35 Agent OS and Boardroom Plug
07:37 Wrap Up

The script covers NotebookLM’s new feature for generating 60-second vertical short video overviews, now rolling out to Google AI Ultra and Pro subscribers and expected to reach free users soon. The creator demonstrates examples and explains that each video is generated from a specific NotebookLM notebook’s research, producing AI images, voiceover, and editing in a hands-off workflow, especially when connected via MCP to an agent operating system. They compare the short-video outputs with longer NotebookLM videos (more slideshow-like) and with alternatives like Open Montage (more cinematic) and a separate Video Agent (preferred for educational videos). Despite video quality being below human-made content, they highlight NotebookLM’s strength as a research-and-learning tool and its one-click outputs (audio, videos, slide decks, mind maps, infographics, flashcards, quizzes, tables, reports). The episode ends by promoting the AI Profit Boardroom for setup guides, trainings, and coaching.

00:00 NotebookLM Shorts Update
00:46 What The Shorts Look Like
01:24 Inside Agent OS Integration
02:42 Quality Check And Tradeoffs
03:00 OpenMontage Comparison
04:24 Video Agent Alternative
04:54 NotebookLM One Click Content Suite
05:56 Shorts Vs Long Form Videos
06:45 Learning And Speed Benefits
08:09 Which Tool To Choose
08:49 Join AI Profit Boardroom
09:27 Community Training And Wrap Up

The video compares Claude Sonnet 5 and GLM 5.2 side by side using one-shot builds (dungeon crawler, raycaster maze, multiple games, a website UI, and a Web OS), noting Sonnet 5 sometimes looks smoother but often feels basic or lacks gameplay, while GLM 5.2 is frequently more interesting and polished, though it can be buggy in some tests where Sonnet 5 wins. It also reviews benchmarks (CursorBench and GaudiBench/Goldy Bench), stating Sonnet 5 scores higher than GLM 5.2 on CursorBench but Opus 4.8 outperforms Sonnet 5, and “Fable 5” leads overall and is expected to return within 24 hours. The creator highlights pricing differences (Sonnet 5 far more expensive than GLM 5.2), GLM 5.2’s open-source and OAuth/agentic compatibility, demonstrates plugging GLM into Claude Code via Agent OS, and promotes the Agent OS and AI Profit Boarding community with tutorials, coaching calls, and support.

00:00 One Shot Showdown
00:18 Dungeon Crawler Test
01:07 Raycaster Maze Faceoff
01:52 CursorBench Rankings
02:28 Opus vs Sonnet vs Fable
03:05 Pricing and Agent OS Setup
04:05 GaudiBench and Context Window
04:53 More Builds Mixed Results
07:11 Games and Visual Quality
07:59 Web UI and Web OS
09:08 Final Verdict and Leaderboards
10:19 Dont Chase Models
10:34 Agent OS Offer and Wrap Up

Anthropic announced that Claude Fable 5 will be redeployed globally starting July 1 after U.S. export controls imposed June 12 shut down access to Fable 5 and Mythos 5 for everyone due to immediate compliance needs. The controls were lifted June 30, with Fable 5 returning across Claude Platform, Claude AI, Claude Code, and Claude Cowork for Pro, Max, Team, and select Enterprise plans, included for up to 50% of weekly usage limits before shifting to paid usage credits after July 7. Mythos 5 access is being restored only to certain U.S. organizations via Project Glasswing following June 26 approval. The shutdown followed a report that Amazon researchers found a method to bypass Fable 5 safeguards; Anthropic says the bypass is now blocked in over 99% of cases and that U.S. Commerce testing agreed the new safeguards are extraordinarily strong, alongside plans for deeper government collaboration and pre-release evaluations.

00:00 Fable 5 Returns
00:45 Global Rollout Details
01:13 Usage Limits and Credits
01:54 Mythos 5 Partial Restore
02:32 Why It Was Shut Down
03:46 Timeline and Big Picture
06:19 Amazon Bypass Explained
08:06 New Safeguards Breakdown
09:11 Industry and Government Frameworks
10:05 Wrap Up and Community Plug

The script explains Hermes’ Mixture of Agents system, which lets you combine multiple models (e.g., Claude Opus 4.8 and GPT-5.5) into a panel and choose an “aggregator” model to fuse their outputs, treating the mix as one virtual model. The presenter demos results like generating a Windows-style OS and building games, showing side-by-side comparisons where the mixture outperforms a single model, and notes they tested 42 builds viewable on Goldie Bench. It describes how MOA 2.0 works under the hood (private analysis by models, then an aggregator writes the final answer and runs tools), switching mixes with /MOA, and claims of benchmark gains (8% over Opus 4.8, 11% over GPT-5.5). Downsides include slower runtime, API reliance, and technical setup, which they simplify via their Agent OS dashboard, also featuring Fusion and Sakana-style panels, automations, memory, and access via their paid community with tutorials and coaching.

00:00 Mixture of Agents Intro
00:33 How the Panel Works
01:27 Bench Tests and Demos
02:13 Side by Side Game Results
03:23 Tradeoffs and Limitations
04:02 Making MOA Easy to Use
04:45 MOA 2.0 Explained
05:28 Fusion and the Bigger Trend
06:34 Stop Chasing Models
07:10 Three Systems and Why They Win
07:37 How to Use It Today
08:19 Agent OS and Community Pitch
09:50 Wrap Up and Goodbye

The video reviews Anthropic’s newly released Claude Sonnet 5, described as more agentic and capable of planning and tool use, but argues it underperforms Opus 4.8 on benchmarks (including agentic coding) while costing more. The creator shares Goldy Bench examples Sonnet 5 generated (a ray caster maze, a broken galaxy orbit test, a synthwave background, and a crypt game), noting some outputs look good but others fail. Side-by-side comparisons show mixed results versus GLM 5.2, with GLM succeeding on tasks Sonnet 5 fails, and tweets highlight negative reception focused on poor token efficiency and pricing. The recommendation is to keep using Opus 4.8, expect Fable 5 soon, and focus on building flexible agent systems that can swap models in and out.

00:00 Sonnet 5 Launch
00:30 Benchmarks vs Opus
01:39 Goldy Bench Demos
02:53 GLM 5.2 Comparisons
04:00 Backlash and Pricing
05:57 Fugu Ultra Showdown
07:20 Why Release This
08:00 Focus on Systems
09:11 Agent OS Pitch
09:48 Final Verdict

The episode reports that U.S. export controls on Anthropic’s Fable 5 and Mythos 5 have been lifted, citing a tweet and Anthropic’s official June 30, 2026 announcement that access will begin being restored tomorrow (likely within 24–48 hours). It notes Commerce Secretary Howard Lutnick’s statement about working with Anthropic over two weeks to analyze and approve Fable 5, and mentions a related comment from White House Chief of Staff Susie Wiles. The host recaps the timeline: access was cut on June 16 due to flagged security concerns, reports suggested a return by late June, and the White House/Commerce/Anthropic announcements arrived rapidly on July 1. They also comment that there’s no news on GPT 5.6, argue Fable 5 is a major step up from Sonnet 5, and emphasize building resilient “agent operating systems” that can swap models and avoid workflow disruption, promoting the AI Profit Boardroom and its Agent OS, training, and coaching.

00:00 Export Controls Lifted
01:12 Global Access Returns
01:29 Sonnet vs Fable
01:55 Why This Matters
02:30 Official Statements
03:41 How It Unfolded
04:12 Previous Shutdown Timeline
04:54 GPT Updates Check
05:28 Focus on Systems
06:39 Diversify Your Workflow
07:00 Agent OS Pitch
07:37 Join the Community

Discover the massive new update to Hermes agent that fixes database concurrency and unlocks scalable multi-agent Kanban boards. Learn how to build a Content Factory that runs multiple agents in parallel without freezing to automate complex SEO and content workflows. This update effectively removes the cap on multi-agent collaboration for faster results.

00:00 - 00:00 - Intro: Hermes Kanban Update
00:40 - How to Update Hermes
01:03 - Scaling Multi-Agent Parallelism
02:03 - Fixing Database Concurrency
03:26 - Building a Content Factory
05:07 - End-to-End SEO Automation
06:24 - Local Models and Token Costs
07:11 - Accessing the Agent OS System

Grok 4.5 has been announced as a new update based on a 1.5 trillion V9 foundation model, with Cursor data included for supplemental training, and is currently in private beta at SpaceX and Tesla. Early evaluations suggest performance close to or possibly exceeding Opus 4.8, and Elon Musk frames V9 as a solid workhorse in the same league as Opus. The script discusses timelines and release expectations (including a 42% July estimate and past teaser-to-launch patterns that often land within two weeks), while warning against hype and “Opus killer” claims. It highlights practical advantages of using Grok via OAuth with an existing Twitter subscription versus expensive API usage, showcases builds made with Grok Build, and argues that users should focus on owning robust agent systems (Agent OS, Hermes workflows, benchmarking, memory, and automation) rather than chasing gated or inaccessible frontier models.

00:00 Grok 4.5 Announced
01:07 Why It Matters
01:39 What We Built
02:03 Beta Status and Hype
03:09 Release Timeline Clues
04:21 Models Are Getting Gated
05:03 Focus on Systems
07:32 Local Benchmarks and Models
08:00 Two Week Release Pattern
08:47 Agent OS Demo
09:37 Join the Community
11:24 Final Takeaways

The video showcases a custom “Claude agent operating system” built as a mission control dashboard that unifies AI agents, CLIs, and one-click workflows for tasks like pulling trending news (Hermes Oracle), generating SEO and social content, images/thumbnails, AI avatar videos, and even music, with daily auto-organization and a personal “memory galaxy” that keeps context updated across workflows. The creator argues this is faster, cheaper (via free/cheap APIs, local models), and more customizable than using Claude directly, and highlights rapid iteration by integrating new models quickly and auto-documenting tests in a website with model comparisons. The system also supports orchestrating multiple AI agents (e.g., executive/CTO roles) via Paperclip, and access is offered through the AI Profit Boardroom community with tutorials, updates, coaching calls, and member success stories.

00:00 Agent OS Overview
00:52 One Click Workflows
01:36 Why Not Use Claude
03:12 Build Fast Document Everything
04:17 Orchestrating AI Teams
06:04 Personalization Memory Galaxy
06:36 Community Proof Results
08:29 Daily Building Mindset
09:53 Join The Profit Boardroom
10:36 Tutorials Coaching Wrap Up
11:41 Final Goodbye

Julian demonstrates a new Hermes Mixture of Agents 2.0 setup that stacks multiple models as a panel of agents inside Hermes Agent, aggregating one prompt across several models into a stronger final output. He says he tested 42 builds with the same prompts and shows a leaderboard where systems like Fusion and Hermes Mixture of Agents rank above single frontier models, with side-by-side comparisons on Goldie Bench versus Claude Opus 4.8. He explains how the Agent OS simplifies running Mixture of Agents without terminal commands, saves outputs in a workspace, and includes tools like chat, talk mode, voice-controlled Hermes Jarvis, and Hermes Oracle. The key message is “don’t chase the model, build the system,” noting you can mix cheaper, free, or local models to outperform top single models, and he points viewers to the AI Profit Boardroom for access, tutorials, and coaching.

00:00 Why This System Wins
01:03 How Mixture Works
02:16 Leaderboard Benchmarks
02:50 Opus Comparison Demo
04:01 Setup and Agent OS
04:39 Built In Agent Tools
05:23 System Over Models
06:50 More Builds and Costs
07:58 Get Agent OS Access
08:30 Community Proof and Wrap

The script discusses a Wall Street Journal–reported claim that China’s ZAI model GLM 5.2 matches or beats Anthropic’s highly restricted Claude Mythos on cybersecurity benchmarks like discovering and exploiting software vulnerabilities, despite Mythos being gated to a small vetted group and Fable 5 being taken down and limited by the US government. It highlights that GLM 5.2 is open-weight, free to download, usable locally, and reportedly about a quarter of the cost per token via API, alongside another Chinese entrant, 360 Security, shipping a cyber AI stack. The speaker argues this weakens US labs’ “model moat,” predicts rapid model turnover, and advises focusing on building an agentic operating system with memory, agents, and workflows that can swap in any model. The episode ends by promoting the AI Profit Boardroom and its Agent OS training and community.

00:00 China Matches Mythos
01:31 Open Weights Shockwave
02:44 What WSJ Reported
03:49 Copying Backdoor Rumors
04:15 China 360 Enters
04:53 Moats And Pricing
05:52 Build Systems Not Models
06:47 Why Mythos Was Locked
08:14 Gating Versus Shipping
08:43 New Era Of Pullbacks
09:37 Outcome First Workflow
10:48 Agent OS In Action
12:18 Never Feel Behind
13:08 Get The Agent OS
13:42 Community And Coaching
14:26 Testimonials And Wrap Up

The video demonstrates an app builder system inside an Agent OS using Hermes Agent to turn a single sentence idea into a working single-page app with minimal clicks. The workflow is: enter an idea, agents classify it and draft a plan, you approve or deny, and a coder agent builds, tests, and fixes bugs before shipping the app to a gallery with live previews, starring, deleting, and rebuilding options. A group team chat can generate personalized app ideas by having multiple models (e.g., Claude, Hermes, Gemini) collaborate while reading an Obsidian “memory galaxy” vault that agents continually update and reference. The system also integrates Minimax for a balance of speed and intelligence and a flat-rate coding plan without token concerns.

00:00 One Click App Builder
00:56 Team Chat Idea Engine
02:09 Approve Ship Gallery
02:41 Minimax Speed Pricing
03:48 Why Multi Model Wins
04:29 Obsidian Memory Galaxy
05:31 Rebuild Fix Iterate
07:24 Step By Step Workflow
07:56 Meditation App Demo
08:30 Get Agent OS Access
09:05 Boardroom Community Tour
09:58 Final Call To Action

The video walks through the Hermes Agent Operating System, a UI-based “mission control” dashboard that organizes Hermes Agent and its tools into multiple mini apps to make it more powerful than using Hermes in desktop or terminal form. It demonstrates voice and chat control (including browser actions), building apps with live previews, and running multiple tabs such as Hermes Oracle to pull trending AI news and auto-draft/publish blog and social content to WordPress. The system also includes a Studio for image/video/voice generation via Minimax or Groq, lead generation and outreach campaigns, a mixture-of-experts panel combining models like Claude Opus 4.8 and GPT 5.5, a workspace to save outputs by API, and a “memory galaxy” that syncs context into an Obsidian vault. It emphasizes owning resilient systems over relying on gated models.

00:00 Hermes Agent OS Overview
00:36 Voice Agent Interface
01:43 Browser Control and App Builds
02:22 Hermes Oracle News Engine
03:19 Studio Media Generation
03:51 Profiles and Live Talk
04:20 Lead Gen and Outreach
05:16 Mixture of Experts Panel
06:17 Workspace and Memory Galaxy
07:14 System First Philosophy
07:55 Automation Boards and Workflows
09:13 Model Integrations and OpenClaw
10:06 Get the Full System
10:57 FAQs and Customization
11:46 Testimonials and Wrap Up

The video demos Qwable 27B Coder, a new open-source local coding model on Hugging Face (updated end of June) built on a Qwen 3.6 27B base, showing projects made locally like a polished animated landing page and working games. It’s run on a Mac Studio (Apple M4 Max, 36GB) using Apple MLX (not available via Ollama) and is integrated into an agent operating system to generate live previews and save builds in a workspace. The presenter says Qwable tops their local leaderboards (Goldie Bench/Cody Bench comparisons) and outperforms recently tested local models like Gemma 4 12B Coder, Quifos 9B, and Onif 1.0 on the same tasks, though it runs noticeably slower than smaller models such as Quifos 9B.

00:00 Meet Qwable 27B
00:56 Demos Landing Pages
01:23 Model Specs Setup
02:04 Benchmarks Versus Locals
02:40 Running On Mac
03:03 Agent OS Live Previews
04:24 Speed Tradeoffs
05:00 Real Task Comparisons
05:49 Local AI Is Accelerating
06:05 Agent OS Features Tour
06:40 Community Courses Pitch
07:28 Wrap Up Thanks

Julian answers recent community questions about his Agent Operating System, a system for linking and automating AI agents with tools like voice control (Hermes Jarvis), Memory Galaxy, Kanban orchestration, and a local agent engine. He explains multiple ways to use a Codex subscription inside Agent OS (Hermes, image generation via Codex OAuth, the Codex tab, and a new Codex plugin in Claude Code), emphasizes focusing on repeatable systems over prompts/models, and shares thoughts on Qwen 3.6 vs Qwen Agent World. He covers connecting Codex/Claude Code to local models via configuration or Ollama, highlights GLM 5.2’s strong performance and open weights, and advises on simplifying Agent OS for teams with hidden UI sections and SOPs. He also suggests Agent OS video automation/Remotion for SaaS promo videos, recommends GLM 5.2 for Ollama-first setups, explains how to update Agent OS via the latest zip/update file, and describes using an Obsidian vault as shared memory for individuals and teams.

00:00 Agent OS Overview
00:48 Using Codex Inside
02:24 Systems Over Prompts
04:01 Qwen Agent World
04:37 Local LLM Connections
05:16 GLM 5.2 Benchmarks
07:08 Sharing With Teams
09:14 AI Video Creation
10:37 Best Ollama Model
11:48 Updating Agent OS
12:27 Memory Setup Choices
13:06 Team Shared Brain
14:24 Wrap Up And Invite

The video tests Ornif 1.0, a new free self-learning local open coding model from Deep Reinforce, showing benchmark comparisons where it outperforms models like Gemma 4 31B and even compares well against larger flagship options. The model can be downloaded via Ollama with a simple command and is integrated into the creator’s agent operating system, including Hermes profiles for agentic workflows. Ornif’s key idea is “self-scaffolding”: it writes its own plans/instructions, generates multiple solution rollouts, gets graded, and improves both planning and coding through reinforcement learning. The creator highlights a gallery of apps, games, tools, and visual outputs made with Ornif, plus local and frontier leaderboards, and demonstrates local offline systems like a Hermes engine, agent Kanban orchestration, and local chat, emphasizing local AI benefits like being free, private, fast, and improving.

00:00 Meet Ornif 1.0
00:34 Setup With Ollama
00:45 What Ornif Is
01:04 Real Build Examples
01:51 Benchmark Reality Check
02:11 Self Improving Framework
03:02 Reinforcement Learning Loop
03:52 Self Scaffolding Explained
05:31 Local Leaderboards Demos
06:21 Agent OS Integrations
07:32 Model Comparisons Use Cases
08:05 Join The Community
09:05 Final Thanks

The video tests Qwythos 9B, a Claude-style creative reasoning model built on a Qwen 3.5 9B base and available on Ollama, showing how to install and run it locally on a Mac for free with no cloud or token costs. The creator wires it into their Agent OS local engine and demonstrates it building several small apps (to-do list, digital clock, Snake game, landing page, calculator), noting it can produce surprisingly solid designs but can also be glitchy or messy. While it advertises a one million token context window, the practical window depends on available RAM and Ollama settings, and it may cut off or load smaller than expected. They compare it with other local models like Qwikboss and Ornif 9B (with Ornif 1.0 seemingly outperforming) and promote the AI Profit Boardroom and Agent OS for running local models and agents.

00:00 Meet Qwythos 9B
00:26 Install via Ollama
00:43 Agent OS Integration
01:39 Apps Built Locally
02:29 Speed and Comparisons
03:26 Why It Works
04:20 Million Token Reality
04:54 Model Sizes and Quantization
05:36 Pros and Cons
06:30 Offline Wrap Up
07:09 Plug Into Any Agent
07:40 AI Profit Boardroom Pitch
08:07 Agent OS Feature Tour
08:57 Community and Coaching
09:24 Final Thanks

The episode answers community questions about the Agent Operating System and how to get the most from it, highlighting Hermes Oracle for news and SEO automation, Hermes Jarvis as a voice-activated agent, and managing all agents and client instances from one interface using separate profiles. It argues that gated frontier models like GPT 5.6 previews and the removal of Fable 5 matter less than building robust systems that can swap models in and out, including using alternatives like Fusion or Sakana Fugu and open-weight local options such as GLM 5.2. The host discusses a high-volume posting strategy with strict quality control, rising demand for local model setups for privacy and cost, automating video production via a video agent and avatar workflow training, using Paperclip for fact-checking and multi-agent teams, Windows support, SEO article generation and deployment, UI improvements to the Kanban board, and examples of AI-assisted game creation, while promoting the AI Profit Boardroom for daily Q&A, tutorials, coaching calls, and the Agent OS zip file.

00:00 Agent OS Overview
01:13 Model Access Concerns
03:16 Posting Strategy Growth
04:41 Local Models for Clients
05:36 Managing Client Agents
07:10 Automating Video Creation
07:57 Fact Checking Workflow
08:36 Windows Install Support
08:54 Hermes Jarvis Voice Agent
10:09 Paperclip Social Team
11:57 Systems Over Models
12:37 Hermes Desktop vs OS
13:28 Using GLM in Claude Code
15:21 SEO Blog Automation
16:15 AI Game Development Demos
18:09 Kanban Board Upgrade
19:58 Join the Boardroom
21:02 Easy Setup Testimonials
22:14 Final Wrap Up

Hermes Agent has released Mixture of Agents (MoA) Presets, allowing multiple AI models to run in parallel and be aggregated into a single stronger answer, similar to systems like Fusion and Sakana Fugu. The script explains this as a workaround for gated or limited-preview frontier models (e.g., GPT 5.6 and Claude Fable 5) by improving performance through a “panel of experts” approach rather than relying on one model. It introduces Hermes Bench, noting claimed reference MoA scores 8% higher than Opus and 11% higher than GPT, and highlights top performance from an Opus 4.8 + GPT-5.5 aggregator. Viewers are shown how to update Hermes, select MoA presets via terminal and dashboard, configure via desktop app or config.yaml, and switch presets with provider commands.

00:00 Hermes MoA Presets
00:43 Why Models Are Gated
02:09 Update and Enable MoA
04:06 Commands and Config
04:49 Top Hermes Bench Combos
05:04 Panel Beats Genius
05:47 Fusion and Alternatives
06:35 Build Systems Not Models
07:34 Agent OS Workflow
08:42 Old Way vs New Way
09:39 AI Profit Boardroom
10:58 Wrap Up and Links

The episode breaks down OpenAI’s GPT-5.6 release, explaining that it’s currently a limited preview available only to about 20 partner organizations via Codex and the API at the request of the US government, with general access hoped for in the coming weeks and possibly restricted by country. It introduces the three new models—Sol (frontier, with Sol Ultra and Sol), Terra (balanced, competitive with GPT-5.5 at roughly half the cost), and Luna (fast, affordable for high-volume work)—and highlights benchmark claims such as Sol Ultra’s 91.9 on Terminal Bench 2.1 while noting other tests where competitors still lead. The speaker argues that chasing new model launches is a mistake and recommends building a model-proof “agent operating system” with a swap layer, multiple models on tap, task routing, and owned memory so models can be swapped in and out with minimal effort, then promotes access to their Agent OS and community.

00:00 GPT 5.6 Overview
00:23 Preview Access Limits
00:42 Meet Sol Terra Luna
01:45 Release Timing Uncertainty
03:12 Benchmarks Breakdown
05:27 Stop Chasing Models
07:54 Model Proof System
08:16 Swap Layer Routing Memory
10:43 Common Objections Answered
12:36 Agent OS Offer
13:22 Community Tour
14:19 Final Call To Actionx

Julian answers recent community questions about his Agent Operating System, a system for linking and orchestrating AI agents with features like voice control via Hermes Jarvis, a Memory Galaxy connected to an Obsidian vault, Kanban orchestration, and a local agent engine. He explains multiple ways to use a Codex subscription inside Agent OS (Codex tab, Hermes with GPT‑4.5 and image generation, OpenCloude integration, and a new Codex plugin inside Claude Code), and emphasizes building repeatable systems rather than relying on prompts or specific models. He discusses Qwen 3.6 vs Qwen Agent World, options for connecting Codex/Claude Code to local models via config changes or Ollama, and why GLM 5.2 performs strongly versus other models while being openly available. He also covers simplifying Agent OS for teams with hidden UI sections and SOPs, faster AI video assembly via Agent OS/Remotion workflows, how to update Agent OS via the latest zip and update file, and creating a shared company “brain” with a shared Obsidian vault.

00:00 Agent OS Overview
00:48 Using Codex Inside OS
02:24 Systems Over Prompts
04:00 Qwen Agent World Talk
04:37 Local LLM Connections
05:16 Why GLM 5.2 Wins
07:08 Sharing OS With Team
09:14 Fast SaaS Video Creation
10:37 Best Ollama Model Setup
11:48 Updating Agent OS Builds
12:27 Memory Galaxy Workflow
13:06 Team Profiles Shared Brain
14:24 Wrap Up Join Boardroom

The script covers two announcements: OpenAI’s GPT 5.6 models Sol, Terra, and Luna are released only in a limited US government–requested preview to a small group of trusted partners, with no confirmed public release date, and Anthropic’s Claude Mythos 5 and Fable 5 are also gated to about 100 companies via a deal referenced from CNBC. The speaker argues that instead of waiting for inaccessible frontier models, users can reach similar or better results using multi-model systems: Hermes Agents’ Mixture of Agents, Sakana Fugu, and Fusion, which combine multiple models and use judging/fusion to improve outputs and benchmark performance (including results shown on Goldy Bench). The episode promotes the AI Profit Boardroom community for access to the Agent OS, tutorials, updates, and coaching.

00:00 Two Big AI Updates
00:11 GPT 5.6 Preview Release
01:11 Sol Terra Luna Breakdown
01:33 Why Models Are Gated
02:10 Claude Mythos Fable News
03:03 Best Workaround Systems
03:12 Hermes Mixture of Agents
05:34 Sakana Fugu Benchmarks
06:28 Fusion Beats Single Models
09:23 Cost Local Model Options
10:13 Stop Chasing Models
11:32 Recap and Access Limits
12:08 AI Profit Boardroom Tour
13:42 Final Thanks

The script discusses reports from Axios and The Information claiming the Trump administration asked OpenAI to stagger GPT-5.6’s release, limiting access to government-approved partners during a preview period due to security and cybersecurity concerns, with two federal offices vetting and approving customers one by one and only a hoped-for wider rollout later. It links this to the “Fable 5” situation where a powerful model was taken down after a few days, suggesting a new reality where frontier US models may be delayed or never broadly released. The speaker weighs pros (safety and vetting) and cons (US competitiveness as Chinese open-weight models like GLM rapidly improve and remain widely available), then advises building workflows as model-agnostic systems so models can be swapped without disrupting work, and promotes an agent operating system offered via AI Profit Ballroom.

00:00 GPT 5.6 Restricted
00:52 Fable 5 Fallout
01:19 Government Approval Queue
03:31 Who Gets Access
04:17 Global Competition Risk
06:18 Build Model Agnostic Systems
07:39 Automation Examples
09:09 Agent OS Pitch
10:01 Local Private Options
10:42 Wrap Up and Outlook

The speaker demos an “agent operating system” dashboard that centralizes and orchestrates multiple AI agents to automate building projects, generating and publishing content, and running workflows. The system includes a group chat for CLIs, an idea-to-implementation pipeline, and tools like Hermes Oracle to pull trending X news, draft posts, and publish SEO blog content directly to WordPress. Hermes Jarvis enables voice-controlled building with transcripts and wake-word mode, while an email outreach SaaS module finds, enriches, validates leads, drafts campaigns, and manages inbox/sent and sending limits. The OS also features model panels like Sakana Fugu and Fusion for multi-model “team” outputs, a workspace to store all builds, a “memory galaxy” knowledge graph that auto-logs activity, and automations for Google Search Console keyword research, avatar videos, and media generation, with an option to get the setup via the AI Profit Boardroom community.

00:00 Agent OS Overview
00:24 Mission Control Dashboard
01:30 Hermes Oracle Trends
02:45 Personalized SEO Publishing
03:39 Hermes Jarvis Voice
04:17 Email Outreach SaaS
05:33 Fugu Boardroom Models
06:48 Local Engine Fusion
08:07 Build Process Benchmarks
09:02 Memory Galaxy Graph
10:42 Automating Daily Workflows
13:34 Community Setup Offer
14:54 Final Wrap Up

The episode reviews recent leaks suggesting Anthropic’s Claude Fable 5 may return after being released on June 9 and pulled three days later, with new signals appearing June 24–25. It cites three “checkable trails” moving in the same timeframe: Claude Code v2.1.190 string changes (removing “purchased separately” and adding a weekly-usage message that implies subscription inclusion), Fable 5 reappearing as a live listing in the Amazon Bedrock model catalog, and AWS Bedrock docs/model cards showing the model lifecycle as active with Bedrock IDs for both “Anthropic Claude Fable 5” and “global” versions. The host notes nothing is confirmed, raises the question of US-only availability, and estimates a possible return by early July/within seven days. The video ends by emphasizing building flexible systems (Agent OS) that can swap models quickly and promoting the AI Profit Boardroom where updates, guides, coaching calls, and community support are provided.

00:00 Fable 5 Return Rumors
00:44 What We Know So Far
01:24 Timeline of Events
02:38 Claude Code String Clues
03:21 Subscription Plan Implications
03:56 Amazon Bedrock Listing
05:13 AWS Docs Model Cards
05:44 Regions and Release Window
06:25 Build Systems Not Models
07:23 Agent OS and Boardroom Pitch
07:57 Wrap Up and Link

The script demos a custom “Agent OS” built around Hermes Agent, showing how it segments multiple agents and workflows into one system. It highlights Hermes Oracle for pulling trending news and generating social posts or SEO-optimized WordPress articles in one click with scheduled daily refreshes; Hermes Jarvis, a voice-activated mode that can run real-time actions like opening websites and providing daily briefings; and a new Outreach Agent that finds leads, enriches/validates emails, manages campaigns, inbox/sent items, dashboards, and suggested outreach sequences using API keys and sending caps. It also covers goal mode with autonomous looping and QC, a connected Memory Galaxy for logged/interlinked knowledge, a Kanban board for multi-profile agent orchestration, NotebookLM asset syncing via MCP, video/SEO engines, Paperclip for team-based agent orgs, and an idea-to-implementation pipeline shared via the AI Profit Boardroom with tutorials, updates, community support, coaching calls, and testimonials.

00:00 Agent OS Overview
00:18 Hermes Oracle News
02:02 Jarvis Voice Control
03:06 Outreach Email Engine
05:08 Goal Mode Autonomy
05:41 Memory Galaxy Brain
06:23 Kanban Team Workflow
07:06 NotebookLM Asset Sync
07:37 Video SEO Loop Tools
08:35 MCP Workspace Studio
08:59 Paperclip Agent Orgs
09:22 Idea To Shipping
09:43 Get The Setup
10:23 Boardroom Tutorials Support
11:17 Build Vs Customize
11:41 Testimonials And Wrap

Julian demonstrates the experimental “Hermes Lead Machine,” a Hermes Agent workflow that generates and enriches leads in one click using the Hunter API. Users can either paste an existing list to enrich or describe the exact prospects they want in plain English (e.g., SEO agencies interested in link-building) and the system finds companies and people, captures domains and emails, enriches company details, verifies deliverability to reduce bounces, scores and filters leads (0–100) for fit, and segments them by status (new, enriched, valid, contacted, replied). It then writes personalized openers and outreach emails and can send through a dedicated inbox, all managed in a dashboard that feels like a SaaS tool. He contrasts this with manual, duct-taped outreach and explains setup via Agent OS/AppHub, API key, and email configuration, plus community support and coaching calls.

00:00 Hermes Lead Machine Intro
00:13 Lead Finder Dashboard Tour
00:38 Hunter API Find and Enrich Demo
01:30 Segmentation and Workflow Benefits
02:35 Find Verify Write Pipeline
03:05 Old Outreach vs New Automation
04:26 Why Anyone Can Use It
05:07 Six Parts Explained
07:21 Setup and Common Questions
08:12 Agent OS Offer and Community
09:25 Wrap Up and Next Steps

Hermes Agent has introduced the Sidekick system, adding an animated pixel pet that reflects an agent’s status at a glance—idle, thinking, running tools, waiting, done, or failed—across the CLI, TUI, desktop app, and the Agent OS dashboard. The pet acts as a personality-driven status indicator to improve peripheral awareness versus spinners and log lines, helping users quickly notice completion or failures. Setup involves updating Hermes, browsing the Pet Dex gallery with `Hermes pets list`, installing a pet (e.g., `Hermes pets install Boba select`), adjusting size with `Hermes pet scale`, swapping anytime, or disabling instantly with `Hermes pets off`; users can choose from nearly 3,000 pets or submit their own, and the sprite cannot affect code or files. The episode also promotes the AI Profit Boardroom for the full Agent OS, additional agents (Oracle, Jarvis, outreach/lead gen), trainings, coaching calls, and support.

00:00 Meet the Sidekick Pet
00:45 How Pet Poses Work
01:41 Why It Matters
02:04 Where It Shows Up
02:45 Update and Install
03:29 Customize or Disable
04:01 Gimmick or Useful
05:12 Quick Recap
05:43 Get the Full Agent OS

The script demonstrates an experimental “Hermes Outreach Engine” built inside a Hermes-based agent operating system to automate email outreach for link building and SEO. It shows an Outreach tab with dashboards for leads, validation, sendability, activity logs, lead generation (finding companies, enriching to real contacts, adding notes, hiding/skipping oversized companies), and campaign management (drafts, sent, bounces, audience selection, previews, draft-mode approvals, pausing/stopping campaigns, and reply tracking). The workflow is described as six parts: define targets, enrich, verify emails via Hunter, filter out large companies, send from a separate agent inbox, and track results. Setup uses Claude Desktop to build the tool and connects Hunter API, a separate inbox, Himalaya for inbox reading/replying, and Google Workspace API, with optional OAuth. The episode ends by promoting access to the agent OS and community training in the AI Profit Boardroom.

00:00 Hermes Email Agent Intro
00:24 Outreach Dashboard Tour
01:19 Lead Generation Workflow
02:06 Campaign Builder Overview
02:44 APIs and Inbox Setup
03:29 Six Step Outreach Engine
05:23 Automation and Safety Controls
06:38 How It Was Built
07:40 Cold Email Objections
08:42 Key Takeaways Recap
09:17 Agent OS All In One Hub
09:42 Join the AI Profit Boardroom
10:06 Community Training and Coaching
10:51 Final Wrap Up

The script showcases an AI “agent operating system” built using Claude Desktop, featuring Hermes Oracle (a web-scanning, breaking-news updater that ranks headlines, links to Twitter posts, and can generate content and automate social media using Groq OAuth) and Hermes Jarvis (a real-time, voice-controlled assistant with conversation history, daily briefings, and a wall mode). It also demos Sakana Fugu Ultra, described as highly capable and topping the creator’s benchmark site, Goldy Bench, including examples of generated interactive content. Additional tools include Google Search Console keyword/page optimization, a one-click video editor that researches topics and creates videos with B-roll and an AI avatar, Memory Galaxy for organizing new context, and a looping agent with a “judge” that iterates until quality targets are met. The creator promotes daily updates, tutorials, files, and community access via the AI Profit Boardroom.

00:00 AI Agent OS Reveal
00:13 Hermes Oracle News Scanner
01:12 Hermes Jarvis Voice Assistant
02:37 Sakana Fugu Builds Anything
03:32 Benchmarks and Frontier Tech
04:57 SEO and Video Automation Tools
05:54 Memory Galaxy Knowledge Hub
06:45 Infinite Loop Quality System
07:42 Agents Replacing Human Work
08:59 Get the System and Community
10:15 Final Wrap Up

The speaker showcases an “agent operating system” built around Hermes, including a real-time voice-activated Jarvis mode, Hermes Oracle for web/news research and content drafting with source links, and a connected “memory galaxy” second brain powered by an Obsidian vault. They explain they run the system locally (not on a VPS) for security, while relying mostly on cloud models via APIs/CLIs/OAuth (e.g., ChatGPT Realtime API, Groq OAuth, Claude via CLI), noting local models typically require high-end RTX GPUs. They describe hosting files locally and in the cloud while keeping the database (Obsidian) stored locally and auto-updated by agents. The episode answers viewer questions, demonstrates automated SEO/article deployment and social content creation from trending news, emphasizes focusing on systems over specific models, and invites viewers to get the ready-made Agent OS inside the AI Profit Boardroom with tutorials, community help, and coaching calls.

00:00 Agent OS Overview
01:13 Local vs VPS Setup
04:40 Automated Content Engine
06:27 Skill Level and Mindset

The video compares two new “panel” APIs—Sakana’s Fugu (Ultra and Mini) and OpenRouter Fusion—both claiming benchmark performance above Fable 5, by testing them side by side on visual/code generations like a landing page, raycaster maze, living spiral galaxy, inner solar system, and other simulations. In these examples, Fugu Ultra consistently produces the best-looking, smoothest outputs, Fusion is generally second, and Fugu Mini is faster but more variable and sometimes buggy. The script notes practical tradeoffs: Sakana/Fugu is cheaper with a flat plan but is heavily limited by daily generation/time caps and token limits, while Fusion is easier to use via API but costs per token. Both are one-prompt systems with long waits, and both are integrated into the creator’s Agent operating system and AI Profit Boardroom community.

00:00 Sakana vs Fusion Intro
00:32 How Panel APIs Work
01:35 Landing Page Showdown
02:17 Maze and Galaxy Tests
03:42 Solar System Quality Gap
04:10 Limits Pricing and Workflow
06:09 Fusion Open World Example
07:07 Fugu Mini Results
08:00 Liquid Simulation Comparison
08:55 Final Rankings and Tradeoffs
09:35 Agent OS and Community Plug
11:12 Wrap Up

Hermes Agent’s “computer use” feature has a new update adding Windows and Linux support and improving macOS, allowing Hermes Jarvis to control a computer by reading the screen and clicking/typing in the background without moving your real cursor or interrupting your work. The script demos opening Obsidian, websites, Google, and writing a note via real-time mode and a more powerful agent mode that can run tasks in the background. It explains the “Hermes Takeover Engine” workflow, emphasizes cross-OS support, and notes you can use different AI “brains” (including local/free options) rather than being locked to one model. Safety is highlighted through approval prompts before potentially destructive actions, plus a stop button. The episode contrasts this with slower, screen-taking tools like Claude Computer Use and promotes the AI Profit Boardroom/Agent OS for setup, training, and support.

00:00 Update Overview
00:16 Live Demo Basics
00:37 Agent Mode Tasks
01:18 Voice Interaction Demo
01:43 Takeover Engine Explained
02:57 Five Step Quiet Control
03:43 Models Setup Commands
05:45 Agent OS Pitch
06:20 Safety Approvals
06:52 Why Hermes Beats Others
08:40 Key Takeaways
09:07 Join And Wrap Up

The script explains why “loop engineering” is replacing manual prompting: instead of iterating with an AI yourself, you define what “done” looks like, let an agent act, and use an independent judge model to verify results and repeat until the goal is met. It cites industry figures and examples (including Anthropic and NVIDIA’s GTC keynote) to argue that better loops, not better prompts, will power future self-iterating agent systems. The speaker describes a doer/judge loop structure and shows four implementations inside an “agent operating system”: a Fusion loop with configurable rounds and separate builder/judge models, an agent Kanban board with planner/builder/reviewer roles, a Fusion “boardroom” combining multiple model outputs via a judge, and a Sakana/Fugu council approach. It also describes an Obsidian-based memory system and promotes the AI Profit Boardroom community and resources.

00:00 Why Loops Beat Prompts
00:48 What Loop Engineering Means
03:39 Prompting vs Looping
04:44 Doer and Judge Framework
05:37 Fusion Loop Walkthrough
06:26 Agent Kanban Loops
07:21 Fusion Boardroom Method
08:31 Sakana Fugu Council
09:05 Agent OS Bundle Overview
10:01 Memory Feedback Loop
11:26 Getting Started and Proof
12:01 Join the Community
12:35 Final Wrap Up

The episode answers recent questions about building and using an agent operating system (Agent OS) to organize multiple AI agents (e.g., Hermes, Jarvis) in one place, automate pipelines from idea to implementation, and deploy outputs like SEO content. It demonstrates using a Kanban board to triage tasks and complete an SEO keyword research report, then explains ways to resolve blocked tasks (commenting, moving cards, asking Hermes to sync and self-unblock, or using Claude to improve the system). It outlines options for agent orchestration via group chat, pipelines, and Paperclip teams, and recommends Agent OS over Hermes Workspace due to sync issues. The script also covers OpenRouter Fusion for high-stakes one-shot answers, local vs VPS hosting for Hermes, switching from local models to CLI/API, enforcing Paperclip output paths via working directory, simplifying Agent OS for clients, and integrating Obsidian memory and NotebookLM into the workflow.

00:00 Agent OS Overview
01:24 Kanban Swarm Setup
02:06 Fixing Blocked Tasks
04:15 Orchestrating Agents Together
05:13 Why Workspace Sync Fails
06:05 OpenRouter Fusion Explained
08:07 Where to Run Hermes
08:24 Game Studio Model Fixes
09:18 Paperclip Output Paths
10:18 Client Ready Agent OS
11:46 Custom Upgrades Showcase
13:31 Best Model Stack Choices
14:15 Obsidian Shared Memory
16:09 NotebookLM MCP Workflow