Overview
Refrain is a self-healing browser automation framework. AI generates reproducible runbooks, reruns execute with zero AI tokens, and AI agents auto-patch your automation when UIs change. Traditional browser automation (Playwright, Selenium) breaks with every UI change. Autonomous AI agents produce different results every time. Refrain combines the best of both:- AI generates a reproducible runbook by exploring your web app
- Zero AI cost on reruns — deterministic execution needs no tokens
- Self-healing — when UIs change, AI agents detect selector drift and auto-patch your runbooks
- Human-in-the-Loop — Slack / Teams / Discord approval workflows keep you in control
Three interfaces
CLI
Generate and execute runbooks from the command line. Ideal for developers and CI/CD pipelines.
SDK
TypeScript SDK for programmatic access. Embed browser automation into your applications.
Web Console
Web-based management UI. Create, schedule, and monitor runbook executions without code.
How it works
AI builds a runbook
AI explores your app, records each action, reviews the steps, and outputs a reproducible runbook.
Execute deterministically
The runbook executes with zero AI tokens. Whether you run it once or a hundred times, the cost stays at zero.
Key features
| Feature | Description |
|---|---|
| AI-built runbooks | Set a goal and AI autonomously builds reproducible browser operations. No code, no manual scripting. |
| Zero rerun cost | Generated runbooks execute without AI tokens. 100% cost reduction on every rerun. |
| Reproducible & auditable | Runbooks are structured artifacts you can version-control, review in PRs, and audit. Same runbook, same result. |
| Self-healing automation | When UIs change, AI agents detect selector drift and auto-patch your runbooks. Multi-layered self-healing keeps your automation running without manual intervention. |
| Human-in-the-Loop | Built-in Slack, Teams, and Discord approval workflows with native HITL governance. |
| BYOK — Bring Your Own Key | Use your own AI keys with Anthropic, OpenAI, Bedrock, Vertex AI, Azure OpenAI, or any compatible endpoint. Your data is never used for model training, and token consumption is dramatically reduced. |
Next steps
Quickstart
Get up and running in 5 minutes.
Plan & Pricing
Compare features across tiers.