Archon Architecture Proof-of-Concept
A three-layer architecture separating identity, runtime logic, and intelligence.
Architecture Overview
┌─────────────────────────────────────────────────────────────┐
│ Layer 1: Hermes Profile (Thin) │
│ - Identity & routing only │
│ - No intelligence or reasoning logic │
│ - Pure configuration (< 50 lines) │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Layer 2: Claw Code Runtime │
│ - All business logic lives here │
│ - Tool registry and execution │
│ - Message routing and orchestration │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Layer 3: Gemma 4 via Ollama │
│ - The actual intelligence │
│ - Local inference at localhost:11434 │
│ - Model: gemma3:4b (available) / gemma4:4b (target) │
└─────────────────────────────────────────────────────────────┘
Files
| File | Purpose |
|---|---|
profile.yaml |
Thin Hermes layer - identity only |
runtime/harness.py |
Claw Code runtime stub |
runtime/tool_registry.py |
Tool definitions and registry |
ollama_client.py |
Gemma 4 interface layer |
test_integration.py |
End-to-end integration test |
Quick Start
# Run integration test
python3 test_integration.py
# Test Ollama connection
python3 ollama_client.py "Hello, Gemma!"
Design Principles
- Profile is THIN - Only identity, no logic
- Runtime owns everything - All intelligence orchestration
- Local inference - No external API dependencies
- Testable - Full integration test coverage
Epic Reference
Implements Epic #370: Archon Architecture Dispatch #371: Proof-of-Concept Implementation