AI News Today | Julian Goldie Podcast

Claude Oceanus Leak: Anthropic’s Mythos-Class Model May Go Public SoonThe script claims a new Anthropic model string, “Claude Oceanus V-1P,” appeared in Anthropic’s console on June 3 and leaked via a Chinese API proxy after red team access was resold, prompting Anthropic to pause the red team group and open an internal investigation. It shares unconfirmed rumors about pricing, speed, and a one-million-word memory, and argues Oceanus is built on the secret “Mythos” model family that Anthropic previously kept locked due to safety concerns. The narrator cites an Anthropic blog post saying Mythos-class models are expected to reach all customers in coming weeks, mentions reports of major productivity gains and Mythos performance claims, and suggests the release could accelerate long-horizon agentic automation. The episode promotes the AI Profit Boardroom and an agent OS to integrate new models when they drop.00:00 Oceanus Leak Shock00:36 Secret Mythos Backstory01:02 How the Leak Happened01:43 Leaked Specs and Pricing02:19 Mythos Going Public Soon02:59 Why Businesses Should Care03:37 Agent OS Pitch04:14 Mythos vs Oceanus Explained04:48 Early Performance Claims05:22 Autonomous Work and Costs06:39 Bigger Implications and Timeline07:33 Final Call to Action08:01 Wrap Up

Call Your Hermes AI Agent on the Phone (ElevenLabs + Twilio) — Full Setup & DemoThe video shows how to call a Hermes AI agent on a phone and have it talk back like a real assistant while still controlling a local setup. Using an ElevenLabs integration connected to Hermes (with Twilio for a callable phone number), the agent can speak and listen over a live phone call and access tools, memory, sessions, and the terminal to do tasks while you’re away. A demo includes writing and editing a poem file in the downloads folder, scheduling a daily Japanese learning session at 3:00 PM, checking prior sessions, and telling a joke. The script explains setup requirements (ElevenLabs API key and Twilio), notes latency depends on model/voice choice (Claude 3.5 Haiku is cited as fast), and mentions an alternative “Hermes Jarvis” talk mode inside an agent operating system plus access to guides and a community via AI Profit Boardroom.00:00 Call Hermes By Phone00:54 Live Phone Demo03:31 What You Can Do04:04 Setup Requirements04:44 How The Call Works05:52 Hermes Jarvis Talk Mode07:05 Speed And Model Tips08:06 Best Use Cases08:54 Get The Full System09:51 Community And Training10:46 Final Wrap Up

Hermes Agent v0.16 Is Here: New Dashboard, Desktop App, Skill Search, Free Models & Voice ControlHermes Agent v0.16 has just released, and this episode walks through what’s changed, how to update from the Hermes dashboard, and how the new unified admin panel lets you manage chat, MCP setup, skills, sessions, models, logs, and scheduled tasks in one place. It also shows the new Hermes Desktop app for Windows and Mac, including connecting to a remote Hermes, managing channels like Telegram or Discord, viewing artifacts, and starting new sessions. Updates covered include simplified Chinese support, a leaner skills setup, NVIDIA Skills integration, and free models available via the portal (Step 3.7 Flash and Nemotron-3 Ultra). The video also demos a voice-controlled “Hermes Jarvis” that can open websites, schedule tasks, and control the computer, plus ElevenLabs API integration for text-to-speech.00:00 v0.16 Release Overview00:13 Update via Dashboard01:30 New Admin Panel Tour02:12 Model Routing Options02:48 Skills Search Install03:44 Desktop App Walkthrough04:14 Why the App Matters04:36 Language Skills Changes04:54 Free NVIDIA Models05:37 Voice Agent Jarvis Demo07:02 ElevenLabs Voice Setup07:45 Wrap Up and Community

Why You Need an Agentic Operating System for Claude (Stop Using AI Like a Chat Box)The script argues that using an agentic operating system with Claude unlocks most of its capabilities by centralizing multiple AI agents, workflows, and shared memory in one customizable dashboard. It contrasts Claude’s standard chat and Projects with an operating system that lets you organize and revisit past work, avoid tab-switching between tools like Claude, ChatGPT, and Hermes, and resume projects across agents via a linked memory system. The speaker highlights rapid integration of new agent features (like Hermes Mission Control) and demonstrates a one-click SEO content pipeline that deploys content to multiple websites, claiming it took about an hour to build. The video promotes the AI Profit Boardroom as a way to get the full setup, updates, tutorials, community support, and weekly coaching calls.00:00 Why You Need Agents00:27 Claude Limits Demo01:05 Projects vs Custom OS01:54 Genius in a Room02:25 Shared Memory System03:27 Fast Integrations04:11 How to Build Yours04:31 One Click SEO Pipeline06:05 Anyone Can Build It06:43 System Improves Daily07:36 New Tools Plugged In08:13 Get the Full Setup08:56 Final Thanks

Hermes Jarvis: Voice-Control Your AI Agent to Run Your Computer & Automate WorkThe script introduces “Hermes Jarvis,” a voice interface that connects a microphone to the Hermes AI agent so spoken commands can trigger real actions on a computer, not just conversation. The narrator demonstrates voice-controlling apps by asking it to create a new note with automation ideas, and scheduling a daily 3:00 PM cron job to research AI automation news and save results to logs and dated JSON files. They explain how it differs from typical voice assistants by being a “doer” that can use tools like web search, file read/write, browser and screen control (opt-in with permissions), and can speak results back aloud. The system uses Minimax M3 voice commands and an NVIDIA Nemotron model, and is available via an Agent OS setup or buildable directly with the right APIs.00:00 Voice Controlled AI Demo01:36 Scheduling Tasks by Voice02:26 Why Most Voice AI Fails02:48 Goldie Jarvis Mode Explained03:49 Build Apps Hands Free04:02 Safety Setup and Models05:31 Recap and Agent OS Tour06:25 Get the Full Setup07:14 Final Thanks

Odysseus vs Hermes AI Agent: Side-by-Side Test (Which One Wins?)The video compares PewDiePie’s new open-source Odysseus AI agent with the open-source Hermes Agent by testing them side by side. Odysseus is presented as a polished, self-hosted standalone app with a fixed UI, built-in tools like email and calendar, and a setup geared toward running local models, but it can feel messy due to pop-up tool windows and has a more manual, non-live memory approach. Hermes is positioned as a more configurable standalone agent that can run in a terminal, desktop app, or within an agent operating system, supports pluggable memory and cloud or local models, can run in the background 24/7, and integrates with other agents and systems. Using the same model in both produced similar general responses, but the creator prefers Hermes for customization, orchestration, and long-term flexibility.00:00 Odysseus vs Hermes00:29 Getting Odysseus Setup00:47 App Design Differences01:32 UI Workflow Issues02:02 Memory and Skills03:00 Solo App vs Orchestration03:46 Local vs Cloud Models04:57 Performance and Customization05:48 Who Each Is For06:42 Future Proofing Agents07:40 Why Custom Systems Win08:53 Verdict and Wrap Up09:07 Boardroom and Resources09:41 Final Goodbye

NVIDIA Nemotron-3 Ultra Is Free in Hermes Agent (Setup + Why It Matters)The script introduces NVIDIA Nemotron-3 Ultra, a newly released open-source 550B mixture-of-experts frontier model now available for free for two weeks through News Portal and usable inside Hermes Agent. It explains that the model is designed for long-running agentic work (planning, tool use, failure recovery) and is advertised as five times faster while using 30% fewer tokens on agent tasks by activating only the needed experts. The walkthrough shows how to enable it via Hermes Agent’s new Mission Control dashboard by going to Manage → Models and selecting Nemotron-3 Ultra, then testing it in chat. It highlights potential benefits for Hermes “goal mode,” mentions benchmark comparisons against GLM 5.1, Qwen 3.5, and others, and notes it’s also available on Hugging Face for local hosting.00:00 Nemotron Free Drop00:37 Mission Control Setup00:59 Why It Matters01:48 MoE Speed Explained02:49 Three Step Install03:25 Real World Tests04:18 Goal Mode Power05:16 Benchmarks Breakdown06:07 Wrap Up Recap06:13 Community Pitch07:06 Final Goodbye

Odysseus vs Hermes AI Agent: Side-by-Side Test (Which One Wins?)The video compares PewDiePie’s new open-source Odysseus AI agent with the open-source Hermes Agent by testing them side by side. Odysseus is presented as a polished, self-hosted standalone app with a fixed UI, built-in tools like email and calendar, and a setup geared toward running local models, but it can feel messy due to pop-up tool windows and has a more manual, non-live memory approach. Hermes is positioned as a more configurable standalone agent that can run in a terminal, desktop app, or within an agent operating system, supports pluggable memory and cloud or local models, can run in the background 24/7, and integrates with other agents and systems. Using the same model in both produced similar general responses, but the creator prefers Hermes for customization, orchestration, and long-term flexibility.00:00 Odysseus vs Hermes00:29 Getting Odysseus Setup00:47 App Design Differences01:32 UI Workflow Issues02:02 Memory and Skills03:00 Solo App vs Orchestration03:46 Local vs Cloud Models04:57 Performance and Customization05:48 Who Each Is For06:42 Future Proofing Agents07:40 Why Custom Systems Win08:53 Verdict and Wrap Up09:07 Boardroom and Resources09:41 Final Goodbye

NVIDIA Nemotron-3 Ultra Is Free in Hermes Agent (Setup + Why It Matters)The script introduces NVIDIA Nemotron-3 Ultra, a newly released open-source 550B mixture-of-experts frontier model now available for free for two weeks through News Portal and usable inside Hermes Agent. It explains that the model is designed for long-running agentic work (planning, tool use, failure recovery) and is advertised as five times faster while using 30% fewer tokens on agent tasks by activating only the needed experts. The walkthrough shows how to enable it via Hermes Agent’s new Mission Control dashboard by going to Manage → Models and selecting Nemotron-3 Ultra, then testing it in chat. It highlights potential benefits for Hermes “goal mode,” mentions benchmark comparisons against GLM 5.1, Qwen 3.5, and others, and notes it’s also available on Hugging Face for local hosting.00:00 Nemotron Free Drop00:37 Mission Control Setup00:59 Why It Matters01:48 MoE Speed Explained02:49 Three Step Install03:25 Real World Tests04:18 Goal Mode Power05:16 Benchmarks Breakdown06:07 Wrap Up Recap06:13 Community Pitch07:06 Final Goodbye

Stop Typing! Build AI Voice Agents with MiniMax M3 & OpenClaw

Learn how to build and interact with real-time AI voice agents using the new MiniMax M3 update for OpenClaw and Hermes. This hands-free workflow allows you to automate tasks and research in real-time without ever touching your keyboard.

00:00 - Intro: Real-Time Voice Chat Demo

01:11 - Voice Agent Mastery Explained

01:57 - Why MiniMax M3 is the Most Agentic Model

02:43 - Testing Deep Voice & Social Search

03:12 - Operating AI Agents from Your Phone

03:37 - The Voice-to-Text-to-Speech Workflow

04:43 - Recap: Hands-Free AI Interaction

05:55 - Advanced Training & Community Access