Show HN: Libretto — Making AI browser automations deterministic
What it does
Saffron Health has open‑sourced Libretto, a toolkit that aims to make browser-based automations more robust and repeatable. It has been reported that Libretto gives coding agents a live browser and a token‑efficient CLI to inspect pages, capture network traffic, record user actions and replay them as scripts, and interactively debug workflows against real sites. The GitHub repository is available at github.com/saffron-health/libretto and the project runs a documentation site at libretto.sh/docs.
How it works
Install with npm install libretto and get going with npx libretto setup — the setup routine reportedly detects provider credentials (OpenAI, Anthropic, Gemini, Vertex) and pins a snapshot model into .libretto/config.json. Libretto can run as a skill for coding agents: allegedly it can read your on‑screen actions, replay them as Playwright TypeScript, and even convert recorded browser flows into direct network requests by analyzing captured traffic. Want to fix a flaky selector? It can reproduce failures against the live site and let an agent pause, inspect, and patch the script.
Why it matters
This hits a nerve. As LLM‑driven developer tools proliferate, the weakest link often remains brittle UI automation. Who wants flaky scrapers and endless selector hunting? Libretto promises a path from brittle UI scripts to faster, more reliable API calls — and, crucially, a debugging loop against the live site. For teams juggling integrations with complex healthcare software, that’s a breath of fresh air. Open‑sourcing it means other teams can borrow Saffron’s playbook instead of reinventing the wheel.
Libretto isn’t magic; it's tooling that leans into trends we already see: agents that act in browsers, model‑backed snapshots, and a push toward deterministic, auditable automations. Try npx libretto status or npx libretto open
Sources: github.com/saffron-health, Hacker News
Comments