Sign inGet started

Livedocs AI

Flexible AI assistant for data work with chat sessions, agent modes, and model choice

Livedocs AI is a flexible way to use AI for data work.

Unlike a typical chat-based assistant, Livedocs AI shows the work — you see the generated queries, code, and transformations, not just an opaque answer you have to trust. You also get to pick your own model — we don’t lock you in.

Chat Sessions & Tagging

Livedocs AI works through chat sessions. You can:

  • Tag context: Use @ to reference cells, variables, tables, files, or datasets in your prompt.
  • Drop files: Drag files into the chat prompt box.

This lets the assistant understand exactly what you’re working with.

What You Can Do with Livedocs AI

1. Fix Errors in Code or Queries

  • When you hit a Python or SQL error, click “Fix with assistant” in the stack trace.
  • In the Problems section, click the bulb icon on an error or warning to get an AI-generated fix.

2. Explain Cell Content

  • Great for onboarding or understanding unfamiliar code.
  • Example: “Explain this SQL query that calculates monthly churn rate.”

3. Build & Explore

  • Ask the assistant to:
    • Train and evaluate models.
    • Explore large files and uncover insights.
    • Fetch and process external datasets.
  • The assistant can also search the web for datasets, documentation, or articles if you say:
    • “Search the web for…”

Prompting Guide

The quality of results depends heavily on the clarity and context you provide.

General Principles

  1. Be specific

    • Bad: “Make a dashboard.”
    • Good: “Create a dashboard showing revenue growth by region, using @sales_data.”

  2. Provide context

    • Mention the goal: “I want to explain churn trends to executives.”
    • Tag relevant resources: @customer_table, @analysis_cell.

  3. Iterate with the assistant

    • It may ask follow-up questions. Give it the most complete answers you can.

  4. Don’t worry about schemas

    • If you tag a table, the assistant already knows the schema and column types.

  5. Use the right model for the job

    • Some models are better at analysis, others at writing explanations. You can switch models anytime.

  6. Leverage the web

    • Add external knowledge by saying: “Search the web for recent papers on time series forecasting.”

Pricing

  • Every plan gets $10 of free AI usage across any model.
  • Paid plans can set higher limits and pay-as-you-go.