Agent Architecture

Understand the internal structure and decision-making process of Claw-Nomad agents.

Last updated: January 15, 2025

Claw-Nomad agents are built on a modular architecture that separates understanding, planning, and execution. This design ensures transparency, safety, and flexibility.

Architecture Overview

┌─────────────────────────────────────────────────────────┐
│                    USER INTERFACE                        │
│  (Dashboard / CLI / API)                                │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                  COMMAND PARSER                          │
│  Natural Language → Structured Intent                   │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                  PLANNING ENGINE                         │
│  Intent → Execution Steps → Permission Check            │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                 EXECUTION RUNTIME                        │
│  Steps → Platform Actions → Results                     │
└─────────────────────────────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│               PLATFORM CONNECTORS                        │
│  Gmail │ Wallet │ Uniswap │ Moltbook │ ...             │
└─────────────────────────────────────────────────────────┘

Core Components

Command Parser

The Command Parser transforms natural language into structured intent:

// Input: Natural language command
"Swap 0.1 ETH for USDC when gas is below 20 gwei"

// Output: Structured intent
{
  "action": "swap",
  "params": {
    "from": { "token": "ETH", "amount": "0.1" },
    "to": { "token": "USDC" }
  },
  "conditions": [
    { "type": "gas_price", "operator": "<", "value": "20 gwei" }
  ],
  "trigger": "conditional"
}

Planning Engine

The Planning Engine converts intent into executable steps while checking permissions:

// Execution plan for the swap
{
  "steps": [
    {
      "id": 1,
      "action": "monitor_gas",
      "platform": "ethereum",
      "params": { "threshold": "20 gwei" }
    },
    {
      "id": 2,
      "action": "get_quote",
      "platform": "uniswap",
      "params": { "from": "ETH", "to": "USDC", "amount": "0.1" },
      "depends_on": [1]
    },
    {
      "id": 3,
      "action": "execute_swap",
      "platform": "uniswap",
      "params": { "quote_id": "{{step_2.quote_id}}" },
      "depends_on": [2],
      "requires_approval": true
    }
  ],
  "permissions_required": ["wallet:write", "uniswap:swap"]
}

Execution Runtime

The runtime executes steps in order, handling errors and logging everything:

•Dependency resolution — Steps execute in correct order
•Error handling — Automatic retry with backoff
•Approval gates — Pause for user confirmation when required
•Audit logging — Every action recorded with reasoning

Agent Memory

Each agent maintains encrypted memory that includes:

Name	Type	Required	Description
`credentials`	`encrypted`	Required	Platform tokens and keys
`history`	`log[]`	Required	Past commands and results
`preferences`	`object`	Optional	Learned user patterns
`context`	`object`	Optional	Ongoing task state

Privacy Note

Agent memory is encrypted with your personal key. Even Claw-Nomad servers cannot read your stored data.