Hermes Wiki

What Hermes Wiki does

The wiki is organized into concept pages grouped as core architecture, memory and sessions, tools and capabilities, performance, security and reliability, and multi-agent. Core pages cover agent loop and prompt assembly, the tool registry and dispatch, toolsets, the prompt builder, the auxiliary LLM client, provider transport and the provider plugin system. The README states that pages were verified line by line against the source code.

Memory pages describe the three-layer design of MemoryStore, MemoryManager and MemoryProvider, FTS5-based session search, context compression and how skills and memory interact. Other pages cover browser and web tools, the execute_code sandbox, voice mode, context references, prompt caching, smart model routing, credential pooling, checkpoints, tool-loop guardrails and a dashboard OAuth gate. A multi-agent page explains five runtime mechanisms: delegate_task, MoA, background review, send_message and the Kanban board. New topics are tagged with the Hermes version that introduced them, such as v0.13.0 and v0.14.

Key features

• Concept pages for the agent loop, tool registry and prompt builder
• Memory architecture and cross-session recall pages
• Skills system page including Curator background maintenance
• Security, credential pool and checkpoint design notes
• Multi-agent page covering delegate_task, MoA, send_message and Kanban
• Version tags such as v0.13.0 and v0.14 on new topics

When to use it

• Reading how Hermes assembles prompts and dispatches tools before writing a plugin
• Understanding Hermes memory and session search
• Studying Hermes security defenses and loop guardrails

Who it is for: Chinese-reading developers who want to understand Hermes Agent internals from the source.

How it fits with Hermes Agent

A documentation project dedicated to Hermes Agent, based on the Nous Research source code.

FAQ