3 Tools That Make OpenClaw Actually Useful

Getting Started with OpenClaw: Enhancing Your AI Agent

Introduction to OpenClaw Limitations

• The speaker introduces OpenClaw, highlighting its limitations such as lack of email functionality, memory retention, and web browsing capabilities.

• A brief overview of the speaker's previous setup using a Mac Mini M4 running OpenClaw as a 24/7 AI personal agent is provided.

Enhancements for OpenClaw

• The video outlines three key enhancements to improve OpenClaw:

• Dedicated email for the agent.

• A reliable memory system for task retention.

• An advanced browser tailored for agents.

Setting Up Agent Mail

• Introduction to Agent Mail, an email client designed specifically for AI agents, avoiding issues with traditional email services like Gmail.

• Instructions on creating an account at agentmail.to are given; it offers a free tier with three inboxes available.

Configuration Steps

• After setting up the inbox, users need to note their inbox name and obtain an API key from the dashboard.

• Guidance on using Cloud Code to implement commands necessary for integrating Agent Mail into OpenClaw is provided.

Using Cloud Code Effectively

• Explanation of why Cloud Code is preferred over direct use of OpenClaw for building systems and tools due to cost efficiency in token usage.

• Two methods are described for adding skills to OpenClaw: manual skill teaching or utilizing Clawub, a repository of skills.

Caution with Clawub Skills

• Users are warned about potential malware risks when downloading skills from Clawub; caution is advised during downloads.

• The installation process of the Agent Mail skill from Clawub is discussed, emphasizing its utility in managing emails through scripts.

Final Setup Steps

• Instructions on installing Claw Hub via npm command are shared; this allows access to various skills including Agent Mail integration.

• The final prompt needed to install the Agent Mail skill through Claude code is mentioned, ensuring users can connect their inbox effectively.

OpenClaw: Enhancing AI Agent Functionality

Introduction to OpenClaw and Agent Mail

• OpenClaw introduces an AI agent that integrates with Telegram, notifying users of new emails in their agent mail inbox.

• Users can utilize agent mail for personal tasks like forwarding invoices or newsletters, receiving summaries and actionable insights from the AI.

• The potential applications extend to business environments, where agent mail could serve as a support inbox handling user inquiries and bug reports.

Memory System Challenges in OpenClaw

• OpenClaw's memory system creates markdown files stored in a local SQLite database but relies on keyword-based searches which can lead to missed information if keywords are not used.

• As interactions increase, the complexity of queries grows, leading to inefficiencies and higher costs due to token usage when retrieving past session data.

Introducing QMD for Improved Memory Search

• QMD is an open-source tool developed by Toby Ludkkey (CEO of Shopify), designed as a hybrid search engine for markdown files that enhances memory retrieval in OpenClaw.

• It combines keyword search with vector search and LLM-powered re-ranking for more accurate results without requiring API keys or extensive local resources.

• Setting up QMD involves following detailed documentation or using Claude Code to automate the installation process.

Utilizing Agent Browser for Web Interaction

• The third tool discussed is Agent Browser, a command-line utility that allows agents to interact with web pages similarly to human users (clicking links, filling forms).

• Unlike Playwright, which is token inefficient during web interactions, Agent Browser operates with 93% less token usage while maintaining functionality.

Installation Steps for Agent Browser

• To use Agent Browser effectively, it must be installed locally on the OpenClaw machine through specific commands provided in the show notes.

• After installation, users should configure Claude Code with prompts that integrate skills from GitHub into their OpenClaw instance for optimal performance.

How Does the Agent Browser Work?

Overview of Agent Browser Functionality

• The agent browser skill allows users to navigate websites and retrieve information programmatically, demonstrated by querying Hacker News for trending topics.

• The process involves taking a screenshot of the webpage and compiling the data into a markdown formatted response, showcasing efficiency with a completion time of 42 seconds.

Advanced Use Cases for Agent Browser

• No API Workflow: Enables interaction with websites lacking APIs by automating data retrieval from web consoles or dashboards without manual browsing.

• Self-Verifying Code: Utilizes Playwright to automate front-end checks, allowing agents to validate code changes through screenshots instead of manual verification.

• Website Monitoring: Monitors dynamic content on websites (e.g., prices, product listings), providing real-time updates on changes.

Security Features and New Skills

• Security Guard Rails: Incorporates measures against prompt injection attacks, enhancing safety when navigating potentially harmful sites.

• Electron Skill: Expands functionality to control Electron applications (e.g., Slack, Notion), broadening the scope of tasks that can be automated.

Enhancements for OpenClaw Setup

• Three tools—Agent Mail, QMD (memory upgrade), and Agent Browser—transform OpenClaw into a more effective assistant.

• Starting with QMD is recommended as it significantly improves user experience; however, integrating all three tools is encouraged for optimal performance.