PTENES
← Track 5 | Module 5.1
Module 5.1

The 6 Claw Architecture Patterns

OpenClaw, ZeroClaw, NanoClaw, TinyClaw, IronClaw, and Agent Zero — when and why to use each standard.

1

What Is a Claw Architecture?

The name “Claw” represents each pattern’s ability to grab a specific set of trade-offs among speed, security, autonomy, and cost.

The 6 Claw Patterns
Standard Focus Complexity Use Case
OpenClawMaximum productivityLowUnrestricted personal use
ZeroClawZero dependenciesLowOffline, Complete Privacy
NanoClawExtreme minimalismMinimumEdge, IoT, Raspberry Pi
TinyClawLocal multi-agentsMediumAgent teams
IronClawZero-Trust SecurityHighCritical environments
Agent ZeroSelf-evolutionHighAutonomous Systems
2

Why Learn the 6 Patterns?

Knowing only one pattern is like having just one hammer. Each Claw solves a different set of problems, and choosing the wrong one can mean weeks of rework.

Advantages of knowing everyone
  • · Informed choice for each project
  • · Smooth migration between standards
  • · Component reuse across Claws
Costs of choosing incorrectly
  • · IronClaw in a prototype = unnecessary slowness
  • · OpenClaw in production = critical vulnerability
  • · NanoClaw for multi-agent use = impossible to scale
3

OpenClaw — The Open Assistant

OpenClaw is INTELECTO's default standard — full access to the file system, internet, and tools. Ideal for personal use when you are the only user.

# OpenClaw config (intelecto.yaml)
architecture: openclaw
security:
  sandbox: false
  file_access: unrestricted
  network: unrestricted
tools:
  auto_discover: true
  require_approval: false
4

ZeroClaw and NanoClaw — Maximum Lightweight Design

ZeroClaw eliminates all external dependencies — works 100% offline. NanoClaw goes further: a single Python file with fewer than 500 lines.

ZeroClaw
  • Model: Local Ollama (Llama 3, Mistral)
  • Memory: Local SQLite
  • Channels: Terminal or local web interface
NanoClaw
  • Size: fewer than 500 lines of Python
  • Dependencies: Only stdlib + one LLM library
  • Target: Raspberry Pi, minimal VPS
5

IronClaw — Maximum Security

IronClaw implements complete Zero-Trust: every operation is audited, tools run in an isolated sandbox, and no action takes place without explicit approval for high-risk operations.

IronClaw Security Stack
5Immutable Audit Log with cryptographic signature
4Human approval for destructive operations
3WASM sandbox for code execution
2Prompt Injection Detection Pipeline
1Fernet encryption at rest and in transit
6

Agent Zero — Self-Construction and Meta-Cognition

The most advanced pattern: agents that analyze their own performance, identify gaps, and create or modify their own tools and memories to improve progressively.

Self-Evolution Loop
Run → Observe → Evaluate → Improve → Persist
Required safeguards
  • · Rate limit for self-modifications per period
  • · Human approval before changes to SOUL.md
  • · Automatic rollback if metrics worsen
← Back to Track 5