Task Handlers

Run your own function instead of the model

A task handler runs your own function in place of the model — for math, lookups, or calling existing code.

Quick Start

1. Run a function instead of the model

   import { Agent, Team, Task } from 'praisonai';
   
   const team = new Team({
     agents: [new Agent({ instructions: 'help' })],
     tasks: [new Task({
       name: 'compute',
       description: 'add the numbers',
       handler: () => 'computed without an LLM'
     })]
   });
   
   const results = await team.start();
   console.log(results[0]); // "computed without an LLM" — the model was never called

2. Read the workflow context

   new Task({
     name: 'compute',
     description: 'add the numbers',
     handler: ({ input, previousResult, currentStep, variables, task }) => {
       return `step ${currentStep} saw ${previousResult ?? 'nothing yet'}`;
     }
   });

What the Handler Receives

The handler is called with one object.

What the Handler Returns

Return a string, or an object for more control.

Publish variables for later prompts

Variables a handler publishes fill {{placeholder}} slots in later tasks.

const team = new Team({
  agents: [member],
  tasks: [
    new Task({
      name: 'gather',
      description: 'gather',
      handler: () => ({ output: 'raw data', variables: { topic: 'whales' } })
    }),
    new Task({ name: 'write', description: 'write about {{topic}}' })
  ]
});

await team.start();
// The second task's prompt becomes "write about whales"

Stop the workflow

new Task({
  name: 'halt',
  description: 'halt',
  handler: () => ({ output: 'stopping', stopWorkflow: true })
});
// Any task after this one never starts.

A throwing handler fails only its task

new Task({
  name: 'boom',
  description: 'boom',
  handler: () => { throw new Error('handler exploded'); }
});
// This task is marked failed; unrelated tasks still run.

How It Works

Best Practices

Related