Hermes Architecture

This branch is the runtime itself: how Hermes turns a model into an agent. The center is a platform-agnostic AIAgent that the CLI, TUI, Desktop, Gateway, ACP, and API all wrap. Understanding the loop, the prompt assembly, and the backends is what lets you treat Hermes as the substrate for a Personal Agent OS rather than a black box.

The agent is platform-agnostic; the entry points are skins over the same loop. Prompt assembly is tiered — stable → context → volatile — so identity is cached and only the volatile tail changes per turn. The terminal backends, not the profiles, are where real isolation lives.

Planned notes

The AIAgent: the platform-agnostic core every entry point wraps
Entry points compared: CLI, TUI, Desktop, Gateway, ACP, API
The agent loop, step by step (and where it can stall)
Prompt assembly by tiers: stable → context → volatile, and why caching matters
Provider resolution: choosing model/provider per profile or task
The tool registry: 70+ tools across ~28 toolsets, and how they're exposed
Sessions: SQLite + FTS5 storage, what's persisted, what's searchable
The six terminal backends and what each isolates
Why this is a good base for a Personal Agent OS
Local vs gateway runtime: what changes when you leave the CLI

Core sources

Hermes — Developer Guide: Architecture — https://hermes-agent.nousresearch.com/docs/developer-guide/architecture
Hermes — Agent Loop — https://hermes-agent.nousresearch.com/docs/developer-guide/agent-loop
Hermes — Prompt Assembly — https://hermes-agent.nousresearch.com/docs/developer-guide/prompt-assembly
Hermes — Session Storage — https://hermes-agent.nousresearch.com/docs/developer-guide/session-storage
Hermes Agent — repository — https://github.com/NousResearch/hermes-agent

⚠️ Verify each /docs/ path before relying on it — some /features/ and developer-guide paths may have moved.

Connects to: Personal Agent Model · Profiles, Actors & Delegation · Tools, MCP & Plugins · Memory & Personal Knowledge