At the core of FRAKTIΛ lies a layered architecture, where every functional aspect of an agent, reasoning, memory, communication and action—is represented as a modular execution unit. These layers are containerized, hot-swappable and orchestrated by the runtime kernel.

This modularity enables domain-specific optimization, fine-grained control and seamless upgrade paths without rebuilding the entire agent.

Primary Execution Layers

Example: Agent Block Chain

{
  "layers": [
    {
      "type": "input",
      "handler": "normalizeText",
      "output": "parsedInput"
    },
    {
      "type": "inference",
      "model": "openai::gpt-3.5-turbo",
      "input": "parsedInput",
      "output": "intent"
    },
    {
      "type": "memory-write",
      "input": "intent"
    },
    {
      "type": "action",
      "executor": "callWebhook",
      "condition": "intent.includes('alert')",
      "params": {
        "url": "https://alert.fraktia.ai"
      }
    }
  ]
}

Advantages of Modular Execution

✦ Granular Debugging – inspect each layer independently.
✦ Upgrade by Component – replace inference logic without touching memory or I/O.
✦ Parallel Execution Ready – layers can be run concurrently in future runtime versions.
✦ Interoperable by Design – mix custom-built and prebuilt modules.

Visual Analogy

Imagine an agent like a Unix pipeline:

[Input Layer] → [LLM Layer] → [Intent Parser] → [Action Layer] → [Voice Output or API Call]

Each part can be modified, replaced or extended without touching others.