Task Loop

​

What is the Task Loop?

The task loop is SpinAI’s core decision-making process that:

• Analyzes user input
• Decides which actions to take
• Executes actions in order
• Determines when tasks are complete

​

How it Works

1. Decision Making The LLM analyzes the input and available actions:

   // LLM receives:
   - User's input
   - Available actions and their descriptions
   - Results from previous actions
   

2. Action Selection The LLM returns a decision:

   interface LLMDecision {
     actions: string[]; // Actions to execute
     isDone: boolean; // Task completion status
     response: string; // Response to user
     reasoning?: string; // Optional explanation
     summary?: string; // Optional summary
   }
   

3. Dependency Resolution SpinAI orders the actions based on dependencies:

   // Example dependency chain
   getCustomerInfo → validateSubscription → createTicket
   

4. Action Execution Actions run in order, sharing state through context:

   // Each action:
   1. Receives the context
   2. Performs its task
   3. Updates the state
   4. Returns the modified context
   

​

Decision Flow

​

State Management

The task loop maintains state throughout the process:

// Initial state
const context = {
  input: "user question",
  state: {},
};

// After actions run
const context = {
  input: "user question",
  state: {
    customerInfo: { ... },
    ticketId: "123",
    response: "I've created ticket #123",
  },
};

​

Logging and Debugging

The task loop provides detailed logging:

ℹ️ Processing request { input: "I need help" }
ℹ️ Planning next actions ["getCustomerInfo"]
ℹ️ Executing action: getCustomerInfo
ℹ️ Task complete { response: "I've found your info..." }

​

Next Steps