Libretto Makes AI Browser Automations Actually Reliable
Browser automation is one of those things that works great in a demo and breaks every Tuesday in production. Libretto, a new open-source toolkit from Saffron Health, takes a different approach: instead of building yet another AI browser agent, it builds the maintenance layer that keeps browser automations working after you deploy them.
The core idea is that browser automations break constantly — sites update their DOM, buttons move, flows change. Libretto gives you AI agents that can reproduce failures, pause workflows at any point, inspect the live page, and fix issues autonomously. When something breaks, the agent figures out what changed and patches the automation. It also does security analysis on network requests to make sure your automation approach won't get blocked or cause issues.
Saffron Health built this for their own healthcare software integrations, where browser automation reliability isn't a nice-to-have — it's compliance-critical. They open-sourced it because every team doing browser automation has the same pain.
It supports OpenAI, Anthropic, Gemini, and Vertex as AI providers, with the model pinned to a config file. The workflow is straightforward: launch a browser, navigate to URLs, take snapshots, and let the AI agent handle the debugging when things go sideways.
Showing up on Hacker News today with 65 points. The pitch of making automations deterministic resonated — most teams have learned the hard way that getting a browser automation to work once is easy, keeping it working is the real job. Libretto is betting that AI can handle the keeping-it-working part.
https://github.com/saffron-health/libretto
← Back to all articles
The core idea is that browser automations break constantly — sites update their DOM, buttons move, flows change. Libretto gives you AI agents that can reproduce failures, pause workflows at any point, inspect the live page, and fix issues autonomously. When something breaks, the agent figures out what changed and patches the automation. It also does security analysis on network requests to make sure your automation approach won't get blocked or cause issues.
Saffron Health built this for their own healthcare software integrations, where browser automation reliability isn't a nice-to-have — it's compliance-critical. They open-sourced it because every team doing browser automation has the same pain.
It supports OpenAI, Anthropic, Gemini, and Vertex as AI providers, with the model pinned to a config file. The workflow is straightforward: launch a browser, navigate to URLs, take snapshots, and let the AI agent handle the debugging when things go sideways.
Showing up on Hacker News today with 65 points. The pitch of making automations deterministic resonated — most teams have learned the hard way that getting a browser automation to work once is easy, keeping it working is the real job. Libretto is betting that AI can handle the keeping-it-working part.
https://github.com/saffron-health/libretto
Comments