Hermes Agent Supports

Dozens of Kkachi Agents separated by project and role

Root Kernel uses Hermes Agent as its main Agent runtime and operates roles, memory, research, and review flows through KAO, KLM, and KRQ.

Dozens of project-scoped and role-scoped Kkachi Agents work by project and role, splitting one human-defined goal into role-specific work pieces and completing it together.

Operating Evidence

A real operating system for running AI like a team

Root Kernel does not use AI Agents only as tools summoned on demand. It separates projects, roles, memory, and review responsibility, then connects each Agent's work piece into one finished deliverable.

Operating principle	How it works	Why it matters
Role separation	Blue, Red, Orange, Grey, Teal, Green, Yellow, and other Agents are separated by responsibility.	Execution, review, user perspective, documentation, and research are not mixed into one Agent.
Project separation	Agents are operated as project-scoped roles with separated context and work boundaries.	This reduces context and memory contamination while keeping each Agent focused on its scope.
Role-based review	Work is checked again through execution, review, user perspective, and documentation criteria.	AI work does not end as a one-off answer. It is shaped into a deliverable through multiple viewpoints.

Technologies

Hermes-based AI Agent operations with roles, memory, and research queues

Dozens of project-scoped and role-scoped Kkachi Agents work by project and role, splitting one human-defined goal into role-specific work pieces and completing it together.

In operation

KAO

Kkachi Agent Organization

A Hermes-based Agent Operating Model with dozens of project-scoped and role-scoped Kkachi Agents.

PlaceholderKAO magpie-bridge collaboration image. Toward one human-defined Epic, magpie Agents with Blue, Red, Orange, Grey, and Teal color accents connect work pieces like a bridge and complete one central deliverable.

In operation / improving

KLM

Kkachi Letta Memory

A plugin, daemon, and CLI project that helps Hermes Agents use Letta memory and LLM Wiki more accurately.

PlaceholderKLM trust-boundary image. Separate Approved Knowledge, Project Memory, and Retrieval Result, showing that an Agent treats approved knowledge differently from temporary search results.

In operation / improving

KRQ

Kkachi Research Queue

A research and knowledge operations queue where Black and Yellow Kkachi Agents collect, reframe, store, and index knowledge.

PlaceholderKRQ knowledge-reuse image. Black/Yellow magpie Agents review scattered sources into a verified knowledge card, then other Agents reuse that card as work context.

KAO

Agent roles and responsibilities separated by color

KAO uses color roles to separate responsibility.

Color	Role	Description
Blue	Project Leader	Leads execution and communicates with the user.
Red	Red Team Reviewer	Critically reviews Blue’s output and provides judgment axes.
Orange	User Advocate	Checks the result from a user perspective.
Grey	Documentation Steward	Reduces document and context conflicts.
Teal	UX/UI Designer	Supports UX/UI judgment for projects with interfaces.
Green	Business Strategist	Works on business plans, marketing, and market strategy.
Yellow	Research Scout	Strengthens next-item research and hypothesis background.
White	General Task Agent	Handles context-free research and general tasks.
Black	Company Committee	Organizes company-level direction, R&R, and project support knowledge.

Delegation

Delegating work to an Agent organization

Root Kernel does not try to automate everything from the start. A human keeps direction, approval, and final judgment while gradually delegating repeated development, review, documentation, research, and usability checks to role-based Kkachi Agents.

Stage	Work pattern	Human role
Human-centered work	Requirements, development, review, documentation, research, and usability checks are concentrated in one person.	Execution volume and review depth are constrained by human bandwidth.
After KAO operation	Blue executes, Red reviews critically, Orange checks user perspective, Grey keeps documents aligned, and Yellow strengthens research background.	The human compares Agent outputs and adjusts direction and priorities.
Human-in-the-loop delegation	Agents perform work while final approval, key product judgment, and validation criteria remain human-led.	The human focuses more on direction, final approval, and important product decisions.