Skip to content
SAMSKARA Docs

AI Sparkle

AI Sparkle is SAMSKARA’s built-in AI assistant, wired into both notebook cells and SQL Workbench.

In a notebook cell

Section titled “In a notebook cell”

Click the Sparkle button on any Python cell to open an inline prompt bar. Describe what you want in plain language — AI Sparkle sees the current cell, live session variables, and relevant ArrowLake schemas, so it reuses variables you’ve already defined instead of re-deriving them. Suggestions appear as a diff you Accept or Reject before anything runs.

In SQL Workbench

Section titled “In SQL Workbench”

Text-to-SQL generation is DuckDB-dialect aware and understands ArrowLake’s catalog structure, so generated queries target the right catalog and namespace. Results appear as a Monaco diff for review before you run them.

Training it on your own patterns

Section titled “Training it on your own patterns”

Settings → AI includes a Documentation Training UI backed by Vanna + ChromaDB. Paste in your own SQL patterns, table naming conventions, or the sm.* API reference, and SQL Workbench’s AI uses it immediately for future generations. Successful notebook runs are also indexed per-org over time, so the assistant picks up your team’s own conventions.

Running it locally

Section titled “Running it locally”

Set AI_BASE_URL to any OpenAI-compatible endpoint, including a local Ollama instance, to keep every prompt and completion inside your own network instead of calling an external API.

Next steps

Section titled “Next steps”