Gemini

geneva.udfs.gemini.gemini_udf

gemini_udf(
    column: str,
    prompt: str,
    model: str = "gemini-2.5-flash",
    mime_type: str | None = None,
    api_key_env: str = "GEMINI_API_KEY",
    version: str | None = None,
) -> UDF

Return a Gemini UDF with the API key captured from the local environment.

The API key is read from os.environ[api_key_env] at call time and serialized with the UDF. On remote workers the key is available without cluster-level env_vars configuration.

Supports both text and binary (e.g. image) columns. For text columns the prompt is prepended to each value. For binary columns the raw bytes are sent as inline data with the given mime_type alongside the prompt. The column type is detected at runtime from the Arrow array; pass mime_type when the column contains binary data.

Parameters:

• column (str) –

  Name of the input column.

• prompt (str) –

  Instruction sent to Gemini alongside each row's value.

• model (str, default: 'gemini-2.5-flash' ) –

  Gemini model identifier (default gemini-2.5-flash).

• mime_type (str | None, default: None ) –

  MIME type for binary columns. Required when the input column contains binary data; ignored for text columns.

  Supported types:

  • Image — image/jpeg, image/png, image/webp, image/heic, image/heif (docs <https://ai.google.dev/gemini-api/docs/vision>_)
  • Audio — audio/wav, audio/mp3, audio/aac, audio/flac, audio/aiff, audio/ogg (docs <https://ai.google.dev/gemini-api/docs/audio>_)
  • Video — video/mp4, video/mpeg, video/webm, video/mov, video/avi, video/x-flv, video/wmv, video/mpg, video/3gpp (docs <https://ai.google.dev/gemini-api/docs/video-understanding>_)
  • Document — application/pdf, text/plain

  Note: inline data is limited to 20 MB per request.

• api_key_env (str, default: 'GEMINI_API_KEY' ) –

  Environment variable that holds the API key (default GEMINI_API_KEY).

• version (str | None, default: None ) –

  Explicit version string for the UDF so that key rotation does not change the UDF hash and trigger a re-backfill.

• Requires –

  pip install 'geneva[udf-text-gemini]'

Returns:

• UDF –

  A UDF instance ready to be registered with a Geneva dataset.

Examples:

Caption images with a one-sentence description:

>>> udf = gemini_udf(
...     column="image",
...     prompt="Provide a 1 sentence description of the scene",
...     mime_type="image/jpeg",
... )
>>> table.add_columns({"caption": udf})

Summarise text documents:

>>> udf = gemini_udf(
...     column="body",
...     prompt="Summarise this document in 3 bullet points",
... )