bmem

bmem - Open Catalog of Web Skills for AI Agents

The bmem project provides an open catalog of reusable web skills specifically designed for AI agents. This innovative tool enables users to search, add, and execute battle-tested browser recipes using any compatible browser that the agent already possesses, such as Playwright MCP or claude-in-chrome.

Key Features

• Searchable Skill Catalog: Effortlessly find reusable browser recipes that enhance the capabilities of AI agents.
• Integration Flexibility: Add these recipes as native skills into your AI agent, regardless of the existing browser framework.
• Privacy First: The bmem command-line interface (CLI) operates as a consume-only tool, entirely free from interaction with your browser's environment. Your cookies and browsing sessions remain secure, as the agent independently runs the skills without passing sensitive data.

How It Works

The operation of bmem revolves around its CLI, which connects with a separate read-only API for accessing the skill catalog. The API provides a lightweight search endpoint and a manifest to guide the location of skill assets. This architecture enables efficient skills registration into the agent's framework while maintaining user privacy.

Usage Examples

Here are some commands to get started with bmem:

bmem search <query>                            # Search the catalog (e.g., "bmem search linkedin people")
bmem search --site linkedin.com                # Filter search results by site
bmem list [--limit 50]                         # Display the entire skill catalog
bmem show linkedin.com/search-people           # Print the skill's SKILL.md content without installation
bmem add linkedin.com/search-people            # Download and register a skill for use
bmem install                                    # Register the bundled bmem meta-skill for setup

To initiate, run bmem install which triggers the registration of a meta-skill, equipping the agent with command surface knowledge and the capability mapping for different skills.

Skill Composition

Each skill is defined in a SKILL.md format, combining YAML frontmatter with a browser-agnostic recipe. It specifies the necessary capabilities and supplies direct selectors while avoiding full DOM reads, thus enhancing efficiency. The capability interaction involves:

1. Navigating to establish a session,
2. Evaluating the skill's extractor through authenticated data fetching,
3. Returning JSON results while ensuring cookie privacy.

Configuration Details

Users can customize the setup through environment variables:

This value-driven tool, bmem, streamlines the integration of web skills for AI agents, enhancing functionality while prioritizing user privacy and ease of use.