Entrypoints
CLI, TUI, Gateway, ACP adapter
Hermes is not just an agent loop with a bag of tools. It is a layered runtime that separates execution, capabilities, state, memory, skills, delegation, and autonomy. This site turns the repository into a guided reading experience with chapters, checkpoints, and a side assistant for focused questions.
CLI, TUI, Gateway, ACP adapter
AIAgent loop, prompt assembly, context compression
tools, MCP, delegation, environments, file/web/browser
session DB, memory providers, skills, cron jobs
先建立 Hermes 的整体地图,理解它为什么不是一个玩具 agent,而是一套完整平台。
读懂 `AIAgent` 的主循环,理解工具调用、预算控制、缓存与中断是如何组合的。
看清 registry、toolset、schema 暴露和 sync/async bridge,理解工具层为什么能扩展。
把 session ledger、built-in memory、provider recall、session search 和 context compression 拆开看。
理解为什么 Hermes 的 skill 更像 procedural memory,而不是代码插件。
重点看 delegation 的上下文隔离、能力裁剪和结果回流,理解 child runtime 的设计。
看 Hermes 如何把定时任务做成自治执行子系统,而不是一个简单定时器。
最后看不同交互入口如何复用同一套 runtime,把 agent 能力接到用户手里。
Each chapter starts with the one system decision that matters, not a dump of symbols.
Source references point you to the files that define the behavior, so you can verify every claim.
Every chapter ends with a choice question to force a mental model, not passive familiarity.
The drawer carries chapter context so you can ask sharper questions without losing your reading flow.