The Corpus detail view lets you edit a specific Corpus. To open it, go to the Corpus Dashboard (Build section of the Agentic Hub) and click on Edit button on that Corpus, available in both Card view and Table view. This view is organized into four tabs: Configuration, Embeddings, Chunking, and Ingestion. The fourth tab, titled Ingestion, defines how the Corpus parses and processes the content of its Resources before they are chunked and embedded.
The Ingestion tab displays the following fields:
- LLM Provider: Selects the multimodal model provider used when visual interpretation is required, such as for high-resolution or LLM-based processing. It applies to the hi_res and llm strategies, for example openai, vertex_ai, or azure.
- Model Name: Selects the visual model used to interpret images, tables, and scanned documents, for example gpt-4.1. For more detail on visual interpretation, see Multimodal RAG.
- Strategy: Defines how documents are processed before chunking. The default strategy, auto, uses the document's full text. The other options are hi_res and llm.
- Structure: Defines how structured or tabular documents should be interpreted during ingestion, for example table.
- Start Page: Sets the first page of the document to ingest.
- End Page: Sets the last page of the document to ingest.
- Password: The password used to open password-protected documents. The value is masked.
- Logo Process: Determines whether detected logos are processed by the visual model.
- Dialogue: Indicates whether the media file contains spoken dialogue, so the system can use transcription or visual summarization as needed.
- Image Prompt: The prompt used by the visual model to analyze images and describe their content.
- Scanned Prompt: The prompt used to extract and transcribe text from scanned document pages.
- Table Prompt: The prompt used to extract and structure tabular data.
- Media Prompt: The prompt used to generate descriptive captions for video frames.
- Whisper Model: Selects the Whisper model variant used for audio transcription, for example tiny, small, medium, or large.
- Merge: Sets how ingested content segments are merged during processing.
- Frame Sampling Rate: Sets how frequently frames are sampled from video files during ingestion.
Each prompt field can be expanded to a larger editor for easier editing.
A Save button is fixed in the bottom-right corner of the view. Click on Save to persist your changes. After a successful update, a Changes saved indicator is shown in the bottom-left corner.
Since version 2026-07.