Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.therefrain.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

Many banking and government portals still rely on traditional web interfaces with no public API. Teams manually log in to check balances, submit applications, or file tax forms — repetitive tasks that consume hours every week. These portals also refresh their UIs frequently, breaking any conventional automation. Refrain handles these portals by generating a self-healing runbook from a single goal description. When the portal’s UI changes, the runbook automatically adapts — no manual selector fixes needed.

​
Example runbook

name: check-bank-balance
url: https://portal.example-bank.co.jp/login
variables:
  - name: username
    source: secrets
  - name: password
    source: secrets
steps:
  - action: input
    selector: "#userId"
    value: "{{ username }}"
  - action: input
    selector: "#password"
    value: "{{ password }}"
  - action: click
    selector: "#loginButton"
  - action: wait
    selector: ".account-summary"
  - action: extract
    selector: ".balance-amount"
    capture: balance
  - action: export
    format: csv
    fields:
      - name: balance
        value: "{{ balance }}"

​
Generate and execute

1

Generate the runbook

Point Refrain at the portal and describe the goal:
npx @refrainai/cli generate -- \
  --url https://portal.example-bank.co.jp/login \
  --goal "Log in, navigate to account summary, and extract the current balance" \
  --context ./context.md \
  --output ./check-balance.yaml
2

Execute the runbook

Run the generated runbook. No AI tokens are consumed:
npx @refrainai/cli execute -- \
  --runbook ./check-balance.yaml \
  --secrets ./secrets.json
3

Schedule periodic runs

Use the Web Console scheduler or your CI/CD pipeline to run the runbook on a schedule.

​
Why this works well

  • No API required — Operates directly on the web interface, just like a human would.
  • Self-healing — Portal UI refreshes don’t break the automation. Selectors are resolved through multiple fallback layers.
  • Approval gates — Add Slack approval before sensitive actions like fund transfers or form submissions.
  • Zero rerun cost — After the initial AI-powered generation, every execution is deterministic and token-free.

​
What’s next

Generate a runbook

Learn how AI builds runbooks from a goal.

Variables and secrets

Securely pass credentials to runbooks.