Agent OS Index
The whole atlas on one page. Thirteen branches across the build phases, from the mental model up to a working Personal Agent OS — local-first, with security and evaluation running through everything.
01 · Mental Model
- Personal Agent Model — chatbot vs copilot vs agent, the observe→decide→act→verify loop, graded autonomy.
02 · Hermes Core
- Hermes Architecture — the AIAgent, entry points, agent loop, prompt assembly, tool registry, sessions, backends.
03 · Identity & Context
- Identity & Context — SOUL.md vs AGENTS.md vs context files, and a voice that pushes back.
04 · Actors & Delegation
- Profiles, Actors & Delegation — the core branch: profile vs subagent, the profile-vs-skill+delegation criterion.
05 · Native Orchestration
- Native Orchestration — use and extend the native Kanban orchestrator; route work to profiles by description.
06 · Skills & Procedural Memory
- Skills & Procedural Memory — SKILL.md, progressive disclosure, and the self-refining learning loop.
07 · Capabilities
- Tools, MCP & Plugins — core tools, toolsets, MCP, plugins, and the read-vs-write trust model.
08 · Memory & Knowledge
- Memory & Personal Knowledge — MEMORY.md, USER.md, session search, and Honcho's shared memory.
09 · Scheduling & Automation
- Schedulers & Automation — cron as first-class agent tasks; a fresh AIAgent per job.
10 · Messaging Interface
- Telegram Agent Interface — Telegram as a control plane, only after everything is validated locally.
11 · Practical Outputs
- Income Workflows — automations that increase capacity to find, evaluate, answer, and produce sellable outputs.
★ · Always Active
- Agent Security & Ops — threat model, real isolation, and the draft-first rule.
- Evaluation & Observability — trajectories, metrics, and evaluating before automating.
Reference
- Reference Registry — the Hermes docs, tools, and sources cited across the atlas.