Persistent memory for browser automation AI agents is addressed by site-memory, enabling agents to learn and retain site-specific knowledge. This significantly reduces wasted time and computational resources for repeated tasks, making AI agents more efficient and cost-effective. It is designed for users of AI agents that perform browser automation tasks.

How It Works

Site-memory automatically learns and updates knowledge after each website visit. This persistent memory allows AI agents to recall site structures and information, enabling them to navigate directly to goals on subsequent visits without re-learning. This approach optimizes performance for stable websites, leading to faster task completion and reduced token usage.

Quick Start & Requirements

• Install: Use npx skills add LittleYier/site-memory or clone the repository and copy the skill to your agent's skills directory (e.g., ~/.claude/skills/).
• Prerequisites: Node.js 22+ is required.
• Browser Tool Options: Compatible with Chrome DevTools Protocol (bundled), browser-use, Playwright MCP, Claude in Chrome, and any other browser tool capable of navigation and interaction.
• Links: github.com/LittleYier/site-memory

Highlighted Details

• On the WebVoyager benchmark, site-memory reduces costs by 70%-90% and speeds up tasks by 4x with no loss in accuracy.
• A hotel booking task on Booking.com was reduced from 35 steps and nearly 8 minutes to 5 steps and 30 seconds on the second run.
• Aggregate benchmark results show a 67% reduction in total commands and a 76% reduction in total time from Round 1 to Round 3 across 50 tasks.
• Stable sites like Allrecipes and Coursera see agents reaching goals directly after a few visits.

Maintenance & Community

No specific details regarding maintainers, community channels, or roadmap were provided in the README excerpt.

Licensing & Compatibility

The project is released under the MIT license, which is permissive and generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The primary benefit is for repeated browser interactions on stable websites; performance gains may be less pronounced on highly dynamic or frequently changing sites. One benchmark site (Apple) showed an increase in commands and time in later rounds, suggesting potential edge cases or areas for further optimization.