724-office-ai-agent

What This Skill Does

The 724-office-ai-agent is a comprehensive, production-grade autonomous agent system designed to operate 24/7 without dependencies on heavy AI frameworks. Built in pure Python, this agent acts as a persistent digital office assistant capable of handling messaging, file management, and task scheduling. It features a three-layer memory system (session, compressed, and vector storage) that allows it to maintain context over long periods, ensuring that user preferences, operational procedures, and past interactions are never forgotten. With 26 built-in tools and support for MCP (Model Context Protocol) plugins, the agent can be extended to perform custom workflows, execute code, and interact with external systems. Its self-repairing diagnostics and cron-based scheduling make it an ideal choice for persistent, low-overhead automation in business or personal environments.

Installation

To get started, clone the repository and set up your workspace. You will need to install the required dependencies: pip install croniter lancedb websocket-client. If you require WeChat silk audio decoding, add pip install pilk. Create the necessary workspace directories (workspace/memory and workspace/files) and configure your config.json file. Ensure you export your sensitive credentials like OPENAI_API_KEY and WXWORK_CORP_ID as environment variables for security. Launch the agent using python3 xiaowang.py and point your webhook to the server.

Use Cases

This agent is perfect for businesses requiring a persistent presence on communication platforms like WxWork. It is highly effective for automating administrative tasks such as scheduling recurring meetings, summarizing long conversation threads, managing project files, and maintaining a centralized knowledge base. Its ability to run local tools means you can automate server maintenance, file cleanup, or data reporting directly from your messaging interface.

Example Prompts

1. "Check the current status of my scheduled tasks and summarize any pending items from the jobs.json file."
2. "Search through my past session memory to find the project requirements we discussed last week for the marketing dashboard."
3. "Summarize the latest 20 messages in this chat and extract the key action items for the team."

Tips & Limitations

For optimal performance, ensure your lancedb vector store is regularly pruned to maintain search latency. While the agent supports runtime tool creation, treat new tools with caution, especially if they involve filesystem access. Because this is a long-running process, monitor your memory usage in the workspace/memory directory. Do not store plain-text secrets in config.json; always prioritize environment variable injection to prevent data leaks.