Skip to main content
Anchor Browser delivers a state-of-the-art 89% Score on the industry-standard benchmark WebVoyager, leveraging browser-use as a core component of the automation capability.

​
The agent task method

Anchor Browser provides within its SDK the agent.task method that enables natural language control over web browsing sessions. This capability allows you to automate complex web tasks without coding the whole flow.
Looking for versioned reusable workflows? See Creating a Task and Run a Task.

​
Code Example

import Anchorbrowser from 'anchorbrowser';

const anchorClient = new Anchorbrowser({
  apiKey: process.env.ANCHORBROWSER_API_KEY
});

const response = await anchorClient.agent.task(
  'Extract the main heading',                     // Required
  {
    taskOptions: {
      url: 'https://example.com',                 // Either sessionId or url is required
      humanIntervention: false,                   // Disable human intervention during task execution (disabled by default)
      detectElements: true,                       // Improves the agent's ability to identify and interact with UI elements
      maxSteps: 40,                               // Maximum number of steps the agent can take
      agent: 'browser-use',                       // browser-use (default), openai-cua, or gemini-computer-use
      provider: 'openai',                         // For browser-use agent only, openai, gemini, groq, azure, anthropic, xai
      model: 'gpt-5',                             // For browser-use agent only, see model list below
      extendedSystemMessage: 'Focus on extracting the main heading from the page',
      secretValues: {                             // Secret values to pass to the agent for secure credential handling
        API_KEY: 'your-secret-key'
      }
    }
  }
);

console.log(response);

from anchorbrowser import Anchorbrowser
import os

anchor_client = Anchorbrowser(api_key=os.environ.get("ANCHORBROWSER_API_KEY"))

response = anchor_client.agent.task(
    'Extract the main heading',                    # Required
    task_options={
        url='https://example.com',                 # Either session_id or url is required
        human_intervention=False,                  # Disable human intervention during task execution (disabled by default)
        detect_elements=True,                      # Improves the agent's ability to identify and interact with UI elements
        max_steps=40,                              # Maximum number of steps the agent can take
        agent='browser-use',                       # browser-use (default), openai-cua, or gemini-computer-use
        provider='openai',                           # For browser-use agent only, openai, gemini, groq, azure, anthropic, xai
        model='gpt-5',               # For browser-use agent only, see model list below
        extended_system_message='Focus on extracting the main heading from the page',
        secret_values={                            # Secret values to pass to the agent for secure credential handling
            'API_KEY': 'your-secret-key'
        }
    }
)

print(response)

​
Structured Output

The AI object can also be used to extract structured data from the browser. This is done by providing a JSON schema to the AI object, which will then return the structured data. The following demonstrates using Zod and Pydantic to utilize the structured output capability. 
import { z } from 'zod';
import { zodToJsonSchema } from 'zod-to-json-schema';

const anchorClient = new Anchorbrowser()

// Define the expected output structure using Zod schema
const outputSchema = z.object({
  nodes_cpu_usage: z.array(
    z.object({
      node: z.string(),                 // Node name
      cluster: z.string(),              // Cluster identifier
      cpu_avg_percentage: z.number(),   // CPU usage percentage
    })
  )
});

// Execute the AI task with structured output
const result = await anchorClient.agent.task('Collect the node names and their CPU average %', {
  taskOptions: {
    outputSchema: zodToJsonSchema(outputSchema), // Convert to JSON Schema
    url: 'https://play.grafana.org/a/grafana-k8s-app/navigation/nodes?from=now-1h&to=now&refresh=1m',
  }
});
console.info(result);

# Define data models using Pydantic for structured output
class NodeCpuUsage(BaseModel):
    node: str                    # Node name
    cluster: str                 # Cluster identifier
    cpu_avg_percentage: float    # CPU usage percentage

class OutputSchema(BaseModel):
    nodes_cpu_usage: List[NodeCpuUsage]  # List of node CPU usage data

# Create task payload with structured output schema
task_payload = {
    'prompt': 'Collect the node names and their CPU average %',
    'output_schema': OutputSchema.model_json_schema()  # Convert to JSON Schema
}

result = anchor_client.agent.task('Collect the node names and their CPU average %',
    task_options={
        'output_schema': OutputSchema.model_json_schema(),
        'url': 'https://play.grafana.org/a/grafana-k8s-app/navigation/nodes?from=now-1h&to=now&refresh=1m',
    }
)
print(result)

​
Secret Values

Securely pass credentials and sensitive data to AI agents during task execution. Secret values are not logged and automatically cleaned up after completion.

Learn more about Secret Values →

​
Available Models For Browser-Use