Skip to main content

Prism AI

Natural language assistant for BigQuery SQL development.

Prism AI Panel

What It Does​

  • Execute tasks - Run queries, analyze results, edit SQL with approval
  • Search schemas - Navigate your BigQuery projects and tables
  • Optimize queries - Reduce costs and improve performance
  • Debug errors - Diagnose and fix SQL problems
  • Generate reports - Data quality, schema documentation, cost breakdowns

Getting Started​

  1. Open the Prism AI panel (Cmd+I or click the chat icon)
  2. Choose your permission mode: Readonly, Approve (default), or Auto
  3. Select your AI model from the model dropdown
  4. Start chatting

To change your AI provider, go to Settings → AI Assistant.

How It Works​

Context Access: Prism AI can access:

  • Tabs you reference with @Tab 1 mentions
  • Tables you reference with @project.dataset.table mentions
  • BigQuery schemas via built-in tools (list tables, get schemas)
  • Query results you share via @results

Data Flow: When you send a message, it goes through Querylab.io's backend to your chosen AI provider. Conversation history is stored both locally (IndexedDB) and on Querylab.io servers for session continuity.

Permission Modes​

Control what actions Prism AI can take using the mode selector in the chat input:

ModeDescription
ReadonlyAI can only read and answer questions, no modifications
Approve (default)AI can propose changes but requires your approval
AutoAI can execute approved tasks automatically

AI Providers​

Configure your AI provider in Settings → AI Assistant:

ProviderDescription
Auto (default)Automatically picks the best model for your task
ClaudeAnthropic's Claude models (Sonnet 4.5, Opus)
GPTOpenAI's GPT models (GPT-5-nano, GPT-4o)
GeminiGoogle's Gemini models (2.5 Flash, 2.5 Pro)

Desktop users can also use CLI tools (Claude Code, Gemini CLI) for direct provider access with your own API keys. See CLI Integration.

Credits​

Prism AI uses a credit-based system. Each message costs credits based on the model used — lightweight models cost 1x, premium models cost more (shown as a multiplier like "7x" in the model selector).

Your plan determines your monthly credit allowance. Usage resets each billing cycle.

AI can make mistakes. Review generated SQL before running.