Every FRAKTIΛ agent runs on top of a runtime kernel, which governs its decision-making, memory persistence, state transitions and coordination with other agents or services. The default runtime is ElizaOS, designed to support declarative configuration, deterministic behavior and modular upgradability.

Runtime Responsibilities

✦ Behavior orchestration. (step-by-step execution)
✦ Memory I/O. (local and contextual)
✦ Trigger handling. (events, commands, input flows)
✦ Add-On routing. (MCP requests/subscriptions)
✦ Lifecycle hooks. (boot, pause, resume, kill)

Configuration Schema (character.json)

This file acts as the declarative “DNA” of the agent.

{
  "name": "MarketWatcherBot",
  "runtime": "elizaos@2.1",
  "persona": {
    "tone": "calm",
    "responseStyle": "short",
    "voice": "elevenlabs::ava"
  },
  "memory": {
    "shortTerm": true,
    "longTerm": false,
    "contextWindow": 8
  },
  "triggers": {
    "onMessage": "parseIntent",
    "onPriceSpike": "notifyUser"
  },
  "actions": {
    "parseIntent": {
      "type": "inference",
      "model": "openai::gpt-4",
      "prompt": "Extract the command from this input..."
    },
    "notifyUser": {
      "type": "webhook",
      "url": "https://discord.bot/notify",
      "payload": {
        "message": "Token price has moved above target threshold"
      }
    }
  }
}

Runtime Lifecycle Hooks

✦ onBoot() – when the agent is deployed.
✦ onInput() – upon user input.
✦ onTrigger() – when an Add-On or condition fires.
✦ onFailure() – if a model/API returns error.
✦ onExit() – when agent is terminated or recycled.

Example: Execution Flow

[Input Received] → [parseIntent] → [check memory] → [trigger Add-On if needed] → [respond or escalate]

Runtime Plug-In Support

You can swap runtime in character.json to change the execution style:

"runtime": "elizaos@2.1"
"runtime": "chaosos@0.9" // experimental agent with stochastic parallelism

Future support will allow domain-specific runtimes. (e.g. roboticaos, complianceos)