🐒 Ming's Note

← Back to Concepts

Agent Knowledge Architecture

A reusable pattern for structuring AI agent systems: separate the execution layer (agent) from the knowledge layer (brain).

The Pattern

Agent (execution)          Brain (knowledge)
─────────────────          ─────────────────
Skills & SOPs              Reference library
Prompt & config            Accumulated examples
Tools & integrations       Learnings & annotations
Session memory             Persistent knowledge

Why Separate Them

  1. Different lifecycle — Agent config changes frequently; knowledge accumulates slowly
  2. Different ownership — Agent is code; knowledge is content
  3. Different backup needs — Knowledge is the strategic asset; agent is replaceable
  4. Reusability — Same brain can serve multiple agents; same agent can switch brains

Implementation Pattern (from gbrain)

GBrain uses two repos: - Workspace repo — Agent configuration, skills, memory, crons - Brain repo — Knowledge base, people pages, meeting notes, content

Both are git repos. Knowledge grows through ingestion. Agent reads from brain, writes back learnings.

Application for One-Person Company

work-profile/
├── agent/              ← Execution layer
│   ├── skills/         ← SOPs for different operations
│   ├── config.yaml     ← Model + provider settings
│   └── cron/           ← Scheduled tasks
│
└── brain/              ← Knowledge layer
    ├── taste/          ← Curated quality examples
    ├── libraries/      ← Reference materials
    └── learnings/      ← Campaign learnings, patterns

Key Insight

The brain is the strategic moat. Agent is replaceable; accumulated knowledge is not. As Garry Tan puts it: "The point of building a 100K-page brain is to use it as a strategic moat. To never lose context."

Related