agent-browser

Sessionful automation maintains separate browser contexts, cookies, localStorage, and history across calls. Per-session browser contexts ensure isolated state, enabling AI agents to handle sign-in flows and data extraction across multiple sites with reusable templates.

Automating sign-in flows and end-to-end web tests is achieved through scriptable control with open, snapshot, interact, screenshot, execute, and close commands. Reusable templates support browsing workflows across multiple sites with optional video recording and proxy support.

Headless browsing and proxy support are available for web scraping and automation workflows. Optional video recording and proxy configuration enable flexible data extraction and testing across multiple sites while maintaining per-session browser contexts.

Data extraction uses snapshot to return short @e refs that drive actions without parsing full DOM. AI agents interact with these compact element references to extract data efficiently, minimizing token usage while maintaining scriptable control over the browsing session.

Scriptable control limitations depend on browser context isolation and snapshot accuracy for dynamic content. AI agents rely on @e refs from snapshots, which may require re-snapshotting when page elements change during interactions, though per-session state remains maintained across calls.