Video summary

Fable 5 Agentic OS is Insane... just watch

Main summary

Key takeaways

Technology

Overview

The video is a walkthrough of “Agentic OS” powered by “Fable 5”, presented as an extremely powerful AI model. It explains how Fable 5 upgrades an existing agent workflow centered around Claude Code + Hermes Agent across five “levels.”

The speaker emphasizes that Fable 5 unlocks new capabilities for doing more work, building faster, and potentially reducing cost/token usage.

Framing: Fable 5 = a “new class of model” (per the subtitles) that upgrades a connected, agentic operating environment.

Core thesis / product framing

  • Fable 5 = a “new class of model” and “world’s most powerful AI model” (as claimed in the subtitles).
  • The goal is to use Fable 5 to massively upgrade a “Claude/Agentic operating system”—an AI environment that coordinates skills, memory, and automations as a connected whole.
  • A repeated theme is “unified memory” and a single, continuous system/location where different AIs (e.g., Hermes vs. Claude/other agents) can share context.

Level 1: Give Claude the “wheel” (add Fable 5 into the OS)

This level focuses on integrating Fable 5 into the Agentic OS within Hermes Agent, including model selection and reasoning “strength” controls.

Integration path / access

  • Mentions that access “inside the Claude app” may be time-limited, then shifts to API access.
  • Uses Hermes Agent as the interface (“harness”) to access the model.

Model configuration

  • In Hermes, the user selects Claude Fable 5.
  • The user sets a strength level: low / medium / high / extra high.

Interactive command example

  • The OS/agent can answer queries like checking the “last thing saved on the desktop.”

Skill discovery / ROI suggestions

  • The agent can scan stored data (e.g., Claude chats + Hermes chats) and propose new skills to add for greatest ROI.
  • The prompt explicitly asks for:
    • “two skills… not currently using… with best ROI”
    • possibly involving external sources

Cost/token caveat

  • Warns that asking Fable 5 to look at “everything” can be token hungry.

Level 2: “One brain” (agentic shared memory + shared knowledge base)

This level focuses on a memory system that connects multiple tools so Claude and Hermes can access the same knowledge.

Unified context / ultimate memory

  • Aims to maintain long-term knowledge such as:
    • goals
    • financials
    • past conversations
    • meeting notes
    • other durable records

Obsidian knowledge base integration

  • Connects to an Obsidian-based local LLM wiki for retrieval of structured writing/notes.
  • Claude Fable 5 can use wiki content in answers (example: “three best principles for making YouTube video intros”).

Memory UX improvement via Fable 5

  • Fable 5 is prompted to redesign the memory interface to be more interactive and clickable.
  • The result shown includes:
    • topics clustered by theme
    • expandable topic threads
    • navigation features (e.g., search bar, topic drilling, frameworks)
  • The UI may also display token spend or conversation/token-related accounting.

Level 3: “Ministry” (orchestrating multiple expert models)

This level describes a coordination pattern inside Hermes Agent called “Ministry of Experts” (or Ministry of Agents).

Orchestrator + sub-experts

  • An orchestrating agent queries three separate models (as listed in subtitles):
    • DeepSeek V4 Pro
    • GLM 5.2
    • ChatGPT 5.5
  • The system aggregates/debates outputs and returns the best response.

Fable 5 used as orchestrator

  • The speaker replaces the previous “king” model (Opus 4.8 mentioned) with Fable 5 as the orchestrator.

Token/cost efficiency

  • Notes that Hermes caches prompts to reduce token cost in the multi-model setup.
  • Mentions OpenRouter and “pay pennies”/very low cost (as claimed).

Example “debate” task

  • The Ministry decides the “best country” for someone in their early 30s who likes coffee and spring.
  • Example verdict shown: Netherlands (Ut(e/possibly Utrecht area)).

Level 4: “Dreaming” (nightly autonomous review and improvement)

This level introduces the idea that the system works overnight and improves based on all available data.

Autonomous “dreaming” loop

  • Fable 5 is prompted to review existing OS data sources, including:
    • Claude code activity
    • automations
    • skills
    • memory files
    • chat logs
    • Hermes integrations
  • A redesigned dreaming system is shown with a better UI and an overview of model/data sources.

Model architecture visibility

  • The video highlights that dreaming provides an organized view of:
    • which models do the dreaming
    • what data sources feed it
    • how results are returned

Productivity framing

  • The emphasis is that a “super intelligent AI” reviews everything and proposes improvements while the user is away.

Level 5: Follow-you interface (“voice/sonar plasma” interactive OS)

This level makes the OS more embodied by adding an interface that can answer questions and operate using UI/voice-like components.

Interactive voice OS / conversational interface

  • Uses Hermes voice intelligence as an example.
  • Then expands to a more interactive UI via Claude Code.

Fable 5 builds the interface

  • The prompt instructs Fable 5 to build an interactive voice operating system overlay on Claude Code OS:
    • answer questions about the platform
    • support clicking/buttons and navigation
    • use lots of data
    • improve the dashboard and code live

Shown behavior

  • A voice-like chat panel appears; the user asks for status and the system returns stats (e.g., messages sent/replied counts).
  • Demonstrates live UI changes (e.g., dashboard theme blue → green) while editing code on the desktop.

Additional workflow/tooling notes mentioned

  • Hermes Agent is repeatedly recommended as the central platform/bridge.

  • Mentions learning resources:

    • “Hermes master class”
    • “Claude Code masterclass” (referenced as linked in the video description)

  • Onboarding/setup wizard

    • Setup identifies skills/systems.
    • It can compute estimated savings using an assumed hourly rate, outputting dollars saved per day estimates.

Reviews / guides / tutorials highlighted

The video includes tutorial-style steps to:

  1. Add Fable 5 into Hermes
  2. Adjust reasoning strength
  3. Run commands to audit saved data and suggest new skills with ROI
  4. Connect a shared Obsidian wiki as memory
  5. Integrate Fable 5 into Ministry of Experts (orchestrator role)
  6. Redesign/enhance dreaming
  7. Generate an interactive voice OS UI

It also promises learning resources: “Hermes master class” and “Claude Code masterclass.”

Main speakers / sources

  • Main speaker: Jack Roberts (explicitly introduced as the video’s host).
  • Primary source of capability: Fable 5 (via Claude/Hermes), along with:
    • Hermes Agent
    • Claude Code
    • multi-model backends such as:
      • DeepSeek V4 Pro
      • GLM 5.2
      • ChatGPT 5.5 (used in the “Ministry” example)

Original video