Track AI spend in a table

You can store data about your bot’s AI spend in a table that automatically updates throughout all your bot’s conversations. This includes cost from any of your bot’s AI usage, including:

• Autonomous Nodes
• AI Cards
• Queries to a Knowledge Base

You will need:

• A published bot

Step 1: Create a table

1. Create a new table named SpendTable.
2. Add two columns to the table:
   • Cost (set the type to Number)
   • ConversationID (set the type to String)

Step 2: Add a Hook to track AI spend

1. Create a new Hook with the type After Turn End.
2. Paste the following code into the Hook:

// Convert spend from nano-dollars to dollars
const llm_spend = metadata.tokens.cost / 1000000000;

// Create a row for the current turn's AI spend
client.createTableRows({
  table: 'SpendTable',
  rows: [
    {
      Cost: llm_spend,
      ConversationID: event.conversationId,
    }
  ]
});

This Hook runs after each turn—that is, after the bot has received a message, processed it, and sent its response. The code snippet above does the following:

• Retrieves the AI spend for the current turn
• Convert the cost from nano-dollars into standard dollars
• Saves the cost and conversation ID to the SpendTable

Step 3: Test your table

Try chatting with your bot in the emulator. After every message your bot sends, your table should automatically create a new row. The AI spend will display in the Cost column:

Your table will now keep a record of your bot’s AI spend for each turn.

Step 4 (Optional): Calculate total AI spend per conversation

Your table now stores the AI spend for each turn—but you can also store each conversation’s total AI spend in a separate table.

1. Create a second table named TotalSpendTable.
2. Add the same two columns as the first table:
   • Cost (set the type to Number)
   • ConversationID (set the type to String)
3. Create a new Hook with the type After Conversation End.
4. Paste the following code into the Hook:

// Finds table rows with the current conversation ID
const tableResponse = await client.findTableRows({
  table: 'SpendTable',
  filter: {
    ConversationID: event.conversationId
  }
});

// Calculates the total AI spend for the conversation
let totalSpend = 0;
for (let i = 0; i < tableResponse.rows.length; i++) {
  totalSpend += tableResponse.rows[i].Cost;
};

// Creates a row in the new table with the total AI spend for the conversation
client.createTableRows({
  table: 'TotalSpendTable',
  rows: [
    {
      Cost: totalSpend,
      ConversationID: event.conversationId,
    }
  ]
});

This Hook runs after the end of each conversation. The code snippet above does the following:

• Retrieves the AI spend for each turn of the conversation
• Sums them up to get the total AI spend for the conversation
• Creates a row in TotalSpendTable with the total AI spend for the conversation

If you’re using an Autonomous Node in your main Workflow, the conversation will only end when it times out. This means you’ll need to wait for the duration of your bot’s inactivity timeout before your table updates with the total AI spend.