Personal Agent Model

Before any Hermes config, the question is what a useful personal agent even is. The goal of this whole atlas is the jump from "I use agents like a chat" to "I have operational actors that work for me" — so this branch builds the mental model that everything else hangs off: the loop, the autonomy ladder, and where a human stays in the loop.

A chatbot answers. A copilot suggests inside your tool. An agent runs a loop — observe → decide → use tools → save state → ask for approval → deliver a verifiable output. A personal operating layer is several of those actors, coordinated, that persist and accumulate context about you.

Planned notes

• Chatbot vs copilot vs agent vs automation vs personal operating layer
• The core loop: observe → decide → use tools → save state → approve → deliver
• What makes an output verifiable (and why that's the real bar)
• Tool use as the thing that turns talk into action
• Memory and continuity: why a stateless chat can't be an actor
• Planning and decomposition vs reactive single-step prompting
• Human-in-the-loop: approval gates, draft mode, and trust boundaries
• Graded autonomy: the ladder from suggest → draft → act-with-approval → act
• Failure modes: loops, hallucinated tool calls, runaway cost, silent wrong answers
• What an "agent runtime" is and why you want one instead of glue scripts
• Designing for "give real value", not for demos
• A personal taxonomy: which of my tasks are agent-shaped at all

Core sources

• Hermes Agent — repository — the runtime this atlas is built on. https://github.com/NousResearch/hermes-agent
• Hermes Agent — docs — concepts, features, and architecture. https://hermes-agent.nousresearch.com/docs/
• Agentic Systems — agent architecture & tool use — TODO (see SOURCES.md); ground the loop and tool-use claims in a primary source before citing.

Connects to: Hermes Architecture · Profiles, Actors & Delegation · Agent Security & Ops · Evaluation & Observability