IngestionParsing

Table Extractor

Preserves tables as tables during ingestion, instead of flattening them into a row of loose numbers.

A price list flattened into prose is worse than useless: the model reads the numbers but loses which column they were in, and answers confidently with the wrong one.

Layout-aware parsing keeps the header row attached to its cells, so a question about the enterprise tier's retention period retrieves the enterprise row rather than the one above it.

Use Table Extractor in your own workspace.

Create a workspace, connect a source, and put this to work in a few minutes.

Book a demo →

Related entries

Whisper Large v3

Speech-to-text for voice notes in chat and for indexing recorded calls and meetings.

SpeechIngestion

Image understanding

Describe, compare or answer questions about an image — attached to a chat turn, or handled by a vision step.

VisionIngestion

Document OCR

Turns scanned pages and photographed documents into searchable text, keeping the reading order.

IngestionParsing