Metorial and the AI SDK make it easy to build agents that can perform complex, multi-step tasks using multiple integrations.

What Are Agentic Workflows?

Agentic workflows allow your AI to:

• Call multiple tools in sequence
• Make decisions based on tool results
• Chain together different integrations
• Accomplish complex tasks autonomously

Basic Multi-Step Example

import { generateText } from 'ai';

await metorial.withProviderSession(
  metorialAiSdk,
  { serverDeployments: ['your-deployment-id'] },
  async session => {
    const result = await generateText({
      model: yourModel,
      messages: [{
        role: 'user',
        content: 'Find emails from Sarah about the Q4 project, summarize them, and add a calendar event for our next meeting'
      }],
      tools: session.tools,
      maxSteps: 10 // Allow multiple tool calls
    });
    
    console.log(result.text);
  }
);

The AI will automatically:

1. Search emails for messages from Sarah
2. Filter for Q4 project-related content
3. Read and summarize the emails
4. Create a calendar event

Controlling Workflow Length

Use maxSteps to control how many actions the agent can take:

maxSteps: 5  // Up to 5 tool calls
maxSteps: 15 // More complex workflows
maxSteps: 1  // Single action only

Example: Research and Report Workflow

const result = await generateText({
  model: yourModel,
  messages: [{
    role: 'user',
    content: 'Research our competitors on GitHub, check their latest releases, and create a summary document in Google Docs'
  }],
  tools: session.tools,
  maxSteps: 15
});

This agent will:

1. Search GitHub for competitor repositories
2. Check their recent releases and commits
3. Analyze the information
4. Create a formatted document in Google Docs

Monitoring Progress

Use streaming to see the agent's progress in real-time:

import { streamText } from 'ai';

const result = streamText({
  model: yourModel,
  messages: messages,
  tools: session.tools,
  maxSteps: 10,
  onStepFinish: (step) => {
    console.log('Completed step:', step.toolCalls);
  }
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

Best Practices

• Start with a clear, specific prompt
• Use appropriate maxSteps for your task complexity
• Test workflows with smaller maxSteps first
• Monitor token usage for long workflows
• Handle errors gracefully with try-catch blocks

Complex Example: Customer Support Workflow

const result = await generateText({
  model: yourModel,
  messages: [{
    role: 'user',
    content: 'Customer John Smith reported an issue with order #12345. Check the order status, review recent support tickets, check for similar issues, and draft a response email'
  }],
  tools: session.tools,
  maxSteps: 20
});

This demonstrates how Metorial enables sophisticated automation by giving your AI access to multiple systems seamlessly.