Hermes Agent Update: Mixture of Agents, /learn, /journey & Fable 5 Setup

Key Takeaways

• Mixture of Agents (/ma): your orchestrator model (Finn uses Opus 48) sends the prompt to reference models — ChatGPT and DeepSeek v4 via OpenRouter by default — then synthesizes their answers into one higher-quality response. If it doesn't work on the first try, it's almost always because OpenRouter API keys aren't configured yet.
• /learn turns anything into a skill: hand it a URL, a tweet, a prompt, or run it right after a completed task, and Hermes automatically creates a reusable skill plus a memory — no manual skill-writing.
• /journey memory graph: renders a visual chart of every skill and memory your agent has built and how they connect — a quick way to see what it has learned about you over time.
• Self-improvement got cheaper: background work like creating memories and skills is now delegated to cheaper models instead of your primary Opus/GPT model, which cuts a major source of Hermes running cost.
• Vibe coding now has full git controls: the Hermes desktop app shows file diffs, lets you make commits, and lets you open pull requests without leaving Hermes — turning it into a viable standalone coding tool.
• Set up a dedicated Fable 5 profile: Fable 5 is now available in Hermes. Because each Hermes profile is effectively its own agent, Finn recommends creating a separate profile just for Fable 5 and reserving it for complex, multi-step builds — not as a daily driver, since it is expensive.

Commands & Code Mentioned

/ma <prompt>      # Mixture of Agents: orchestrator + ChatGPT + DeepSeek v4 (via OpenRouter), synthesized answer
/learn <url|prompt>   # Turn a tweet, URL, or just-completed task into a reusable skill + memory
/journey             # Open the skill & memory graph for the current agent

Prompt Finn uses to create the Fable 5 profile: "Please make a new Hermes profile called Oracle that uses the Fable 5 model from Claude. Let me know when that's set up so I can use it in Telegram."