Tips for Effective Interaction

This help topic describes some ways to ensure effective interactions between Graph Studio and your AI assistant.

• Just describe what you want. The AI discovers skills and tools automatically — you don't need to know tool names or skill IDs.
• Create → Activate → Query. Always follow this order for graphmarts. After structural changes, reload before querying.
• Shared context is the key to skills. Move repeated instructions into a skill's shared context field. The AI receives it once instead of once per tool.
• Keep tool summaries concise. One line — they help the AI decide which tool to resolve, not explain how to use it.
• Use the REST API for automation. Define skills in JSON files and use curl to add/remove them programmatically — see the Import and Export section here.
• Changes are live. After editing skills, the catalog updates automatically. No restart required.
• Graph Lakehouse references. When activating a graphmart, specify the server by name if there are multiple (e.g., "activate on LocalGQE").
• File references. When asking the AI to load files, specify the file connection, path, and format.
• Results may vary depending on the LLM provider and model configured for your MCP client.