Input processors

Transform Task inputs before the model runs.

An input processor transforms one input before the Task prompt is rendered. Processors belong to a Task revision, so changing them creates a new revision and leaves previous revisions unchanged.

Each entry in input_processors contains:

FieldContract
param_nameA {placeholder} present in the revision’s user prompt.
input_processorA processor name returned by GET .../input_processor.
configOptional processor-specific settings.

Revision validation rejects a param_name that is absent from the user prompt and more than one audio_transcriber. Processor names and configurations must match the registered processor contract before processing can complete.

Available processors

ProcessorInputResult
url_fetcherURL stringReplaces the URL with fetched page content.
document_content_extractorUploaded documentReplaces the target parameter with extracted text.
audio_transcriberUploaded WAV, MP3, AIFF, AAC, OGG, or FLAC fileReplaces the target parameter with a transcription.
perplexity_searchSearch query stringReplaces the query with ranked web results.

The catalog endpoint returns a plain array with each processor’s name, description, and config_schema. Use the returned schema instead of assuming that all processors accept the same settings.

document_content_extractor currently requires the uploaded file’s name in config.filename, although its catalog entry does not publish a configuration schema. Only one audio_transcriber can be configured on a revision.

Files and direct model input

Document extraction and audio transcription consume files supplied with a Task run. File-capable models can instead receive supported image or PDF files directly through the revision’s file_input_mode; an input processor is only needed when the file must be converted before the model runs.

Run records expose input_processor_timing. Processor charges, when present, are reported separately in input_processor_charged_credits and included in the run’s total charged credits.