Cross-Application Workflows

Most enterprise work involves copying data between applications that do not talk to each other: pulling a record from the CRM, entering it into a spreadsheet, filing it in an internal tool, sending a confirmation email. These workflows run on human operators clicking between windows all day.

Northstar operates the desktop the same way those operators do, but 24/7, across thousands of instances simultaneously.

Move data between a web app and a spreadsheet

crm_to_spreadsheet.py

from tzafon import Lightcone

client = Lightcone()

for event in client.agent.tasks.start_stream(
    instruction=(
        "Open Firefox and go to https://crm.example.com. Log in with 'ops@example.com' / 'ops123'. "
        "Navigate to the Contacts page. Find the contact 'Acme Corp'. "
        "Note their email address, phone number, and account status. "
        "Then open LibreOffice Calc. "
        "Enter the headers 'Company', 'Email', 'Phone', 'Status' in row 1. "
        "Enter the Acme Corp data in row 2. "
        "Save the file as /tmp/acme-export.csv."
    ),
    kind="desktop",
    max_steps=40,
):
    print(event)

crm_to_spreadsheet.ts

import Lightcone from "@tzafon/lightcone";

const client = new Lightcone();

const stream = await client.agent.tasks.startStream({
  instruction:
    "Open Firefox and go to https://crm.example.com. Log in with 'ops@example.com' / 'ops123'. " +
    "Navigate to the Contacts page. Find the contact 'Acme Corp'. " +
    "Note their email address, phone number, and account status. " +
    "Then open LibreOffice Calc. " +
    "Enter the headers 'Company', 'Email', 'Phone', 'Status' in row 1. " +
    "Enter the Acme Corp data in row 2. " +
    "Save the file as /tmp/acme-export.csv.",
  kind: "desktop",
  max_steps: 40,
});

for await (const event of stream) {
  console.log(event);
}

Northstar opens the CRM in a browser, finds the data, switches to a spreadsheet application, enters it, and saves the file. Two applications, one instruction.

Process a batch across systems

# Process each record: look it up in system A, enter it in system B
records = [
    {"name": "Acme Corp", "action": "renew"},
    {"name": "Globex Inc", "action": "upgrade"},
    {"name": "Initech LLC", "action": "cancel"},
]

for record in records:
    print(f"Processing {record['name']}...")

    for event in client.agent.tasks.start_stream(
        instruction=(
            f"Open Firefox and go to https://portal.example.com. "
            f"Search for '{record['name']}'. "
            f"Click on their account. "
            f"Click the '{record['action'].title()}' button. "
            f"Confirm the action. "
            f"Report whether it succeeded."
        ),
        kind="desktop",
        max_steps=20,
    ):
        print(f"  {event}")

const records = [
  { name: "Acme Corp", action: "renew" },
  { name: "Globex Inc", action: "upgrade" },
  { name: "Initech LLC", action: "cancel" },
];

for (const record of records) {
  console.log(`Processing ${record.name}...`);

  const stream = await client.agent.tasks.startStream({
    instruction:
      `Open Firefox and go to https://portal.example.com. ` +
      `Search for '${record.name}'. Click on their account. ` +
      `Click the '${record.action}' button. Confirm the action. ` +
      `Report whether it succeeded.`,
    kind: "desktop",
    max_steps: 20,
  });

  for await (const event of stream) {
    console.log(`  ${JSON.stringify(event)}`);
  }
}

Use persistent state for multi-session workflows

Install software, configure the environment once, and reuse it across runs:

# First run: set up the environment
with client.computer.create(kind="desktop", persistent=True) as computer:
    # Install tools Northstar will need
    client.computers.exec.sync(computer.id, command="apt-get install -y libreoffice")
    saved_id = computer.id
    print(f"Environment ready: {saved_id}")

# Subsequent runs: reuse the environment
for event in client.agent.tasks.start_stream(
    instruction="Open LibreOffice Calc, load /tmp/data.csv, and add a 'Total' column that sums columns B through D",
    kind="desktop",
    environment_id=saved_id,
):
    print(event)

// First run: set up
const computer = await client.computers.create({
  kind: "desktop",
  persistent: true,
});
await client.computers.exec.sync(computer.id!, {
  command: "apt-get install -y libreoffice",
});
const savedId = computer.id!;
await client.computers.delete(savedId);

// Subsequent runs: reuse
const stream = await client.agent.tasks.startStream({
  instruction: "Open LibreOffice Calc, load /tmp/data.csv, and add a 'Total' column that sums columns B through D",
  kind: "desktop",
  environment_id: savedId,
});

for await (const event of stream) {
  console.log(event);
}

Why Northstar for cross-app workflows

These workflows are hard to automate traditionally because:

• No APIs. Many enterprise apps, especially legacy ones, have no APIs, or the APIs do not expose what you need
• Multiple applications. The workflow spans two to five systems, each with its own interface
• Human judgment. The operator reads one screen and decides what to enter in another

Northstar handles all three. It operates the desktop like a human, switching between apps, reading data from one and entering it in another, and it runs unattended.

See also

• Tasks: run workflows with natural language instructions
• Shell commands: combine GUI automation with CLI tools
• Lightcone OS: install the software your workflows need
• Computers: persistent state for long-running workflows