Saffron Health released Libretto, an open‑source toolkit that lets coding agents generate deterministic browser automations instead of relying on opaque runtime prompts. By recording manual actions and extracting network traffic, the system produces readable Playwright scripts that can be inspected, versioned, and debugged. A short demo showcases the workflow from user interaction to fully scripted automation in regulated environments.

Existing solutions such as Browseruse and Stagehand depend on custom DOM parsing, which frequently fails on legacy healthcare portals systems and inflates API costs through repeated AI calls. Libretto sidesteps these issues by using a hybrid approach: it couples Playwright UI automation with direct network/API requests captured from the live browser session, improving reliability and evading bot detection.

Developers invoke Libretto via an npm package and a token‑efficient CLI that handles Chromium setup, provider configuration, and session management. Commands such as `libretto snapshot` and `libretto exec` let teams record page states, generate code, and step through failures without touching production data. The project now lives on GitHub under the saffron-health organization, offering a reproducible path for maintaining complex web integrations today.