Task

What is a Task?

A Task in Axiomkit is any asynchronous operation that takes time to complete. This includes things like:

• Fetching data from external APIs
• Writing or reading from a database
• Performing computationally intensive logic
• Sending emails, notifications, or webhook calls

Tasks are automatically managed by the Axiomkit runtime to ensure efficient, concurrent execution without overloading external systems.

Why Task Management Matters

Without task management, executing multiple long-running operations sequentially leads to performance bottlenecks. For example:

// User asks: "What are the current stock prices for 5 companies?"
// Agent needs to call weather API 5 times
// Without task management:
await fetchStockPrice("AAPL"); // 500ms eg
await fetchStockPrice("GOOGL"); 
await fetchStockPrice("AMZN");
await fetchStockPrice("MSFT");
await fetchStockPrice("TSLA");

This would take ~2.5 seconds if each API call takes 500ms.

With AxiomKit’s built-in task manager, calls are scheduled concurrently with safe limits:

// With task management: AxiomKit executes a limited number of tasks concurrently, 
improving response time.
// Runs 3 operations at the same time (concurrent)
// Queues the rest until the first ones finish
// Approximate total duration: 1 second

const results = await Promise.all([
  fetchStockPrice("AAPL"),  // Starts immediately
  fetchStockPrice("GOOGL"), // Starts immediately
  fetchStockPrice("AMZN"), // Starts immediately
  fetchStockPrice("MSFT"), // wait at a queue
  fetchStockPrice("TSLA"), // wait at a queue
]);

Only a limited number of tasks run at once; the rest queue until slots free up.

How Tasks Are Handled in AxiomKit

Every action handler you define is automatically treated as a task:

const fetchStockPriceAction = action({
  name: "fetch-stock-price",
  description: "Retrieve the latest stock price for a company",
  schema: z.object({ symbol: z.string() }),
  handler: async ({ symbol }) => {
    const res = await fetch(`https://api.stocks.com/price?symbol=${symbol}`);
    return await res.json();
  },
});

When the LLM issues multiple <action_call>s, AxiomKit:

• Runs a limited number in parallel
• Queues excess tasks
• Manages retry logic automatically.

Configuring Task Concurrency

Use TaskRunner to control concurrency at the agent level:

Summary

• Action handlers are executed as managed tasks
• Concurrency is configurable per agent instance
• Async best practices prevent bugs and ensure performance
• Tasks are queued, retried, and cancellable by default

The Axiomkit task system ensures your agents scale safely, even in environments with external rate limits or heavy load.