Skip to main content
Plot Plugins make specialized computations available inside Plot SQL. A Plot owns the query and its resulting table; a Plot Plugin owns a reusable operation that the query can invoke. The first-party Plot Plugins cover the two model-backed operations available today:

Ponder

Classify, extract, and generate typed values with one model call per input row.

Embed

Turn text into comparable vectors for retrieval and similarity workflows.

​
How Plot Plugins fit

  1. A Plot selects and shapes Workspace data with SQL.
  2. A Plot Plugin evaluates its operation for the rows that reach the function.
  3. DataFusion returns Arrow values as part of the query result.
  4. A materialized Plot publishes that result in the Parable Pool.
Run the SQL through an SDK for the ordinary path, or use Flight SQL when you are building a protocol-level client. See Query with the SDKs and Flight SQL.
A Plot is a Piece with its own query and result. A Plot Plugin is a function that a Plot can call; it is not another Piece.
Internal authors can implement another operation from the Custom Plot Plugins reference.

Ponder

Use the first-party Ponder Plot Plugin to return typed model output from SQL.

Embed

Use the first-party Embed Plot Plugin to create comparable text vectors from SQL.