Summary 7/24 Office is a self-evolving AI agent system engineered for 24/7 production. Built in ~3,500 lines of pure Python with minimal dependencies, it offers a framework-free, debuggable platform for complex agentic tasks, targeting users needing autonomous, adaptable AI agents.

How It Works This system bypasses common AI frameworks, relying on Python's standard library plus croniter, lancedb, and websocket-client. Its core features a Tool Use Loop with OpenAI-compatible function calling and automatic retries. Memory spans three layers: session history, LLM-compressed long-term storage via LanceDB vector retrieval, and active recall. An MCP/Plugin System enables hot-reloading via JSON-RPC, and runtime tool creation (create_tool) allows dynamic generation of new Python tools, promoting transparency.

Quick Start & Requirements Clone the repo, copy config.example.json to config.json, and populate API keys. Install dependencies: pip install croniter lancedb websocket-client (pilk optional). Create workspace directories (mkdir -p workspace/memory workspace/files) and optionally configure personality files. Run via python3 xiaowang.py, which starts an HTTP server on port 8080 for webhook integration. Docker is needed for the multi-tenant router.

Highlighted Details

• Production-ready, 24/7 operation, developed solo with AI co-development tools.
• Self-evolving