Workflow templates

Templates provide connected processing steps. Choose one for your task, load it onto the canvas, and configure your sources, models, and output location.

Select and use a template

1. On the workflow creation page, choose the option to start from a template.

2. Search by name or purpose, such as knowledge base or information extraction.

3. Select a template to load its steps onto the canvas.

4. Review node settings and fill in source and target resources in the run form.

5. Save the parameters, start a manual run, and inspect node outputs and saved results.

For your first file-processing run, use a small sample. Memory Governance is session-triggered; see Memory Governance.

Choose by task

Template	Purpose
Prepare Document Knowledge Base	Write document text to a knowledge base
Prepare Document Knowledge Base with Image Index	Build both text and image indexes
Document Information Extraction	Parse documents and save extracted fields
Extract Information to Table	Extract fields into a table
Drawing Parsing	Parse drawings and build text and image indexes
Multimodal Information Extraction	Extract fields directly from PDF pages
Dataset Generation	Generate training or evaluation samples from rules
Speech Transcription Q&A	Transcribe audio and build a knowledge base index
Video Transcription Q&A	Transcribe video speech and build a knowledge base index
Memory Governance	Filter sessions, extract information, and save memories

Memory Governance requires the corresponding feature to be enabled. Follow its page after the creation entry becomes available.

Choose between similar templates

Task	Choice and difference
Retrieve document text	Document knowledge base preparation builds a text index and saves parsed files
Retrieve text and images	The image-index variant adds visual manifests and image vectors with matching model settings
Analyze drawing regions and parts	Drawing Parsing analyzes structure before building text and image indexes
Extract fields to files	Document extraction parses first; multimodal extraction reads PDF pages directly and saves JSON
Write fields to a database	Table extraction adds a writing step and requires compatible columns
Retrieve spoken content	Speech and video templates index transcripts; video does not analyze individual frames
Generate training or evaluation material	Dataset Generation needs rules and a schema, followed by sample review
Retain session memories	Memory Governance uses synchronized sessions, filtering, extraction, deduplication, and writing

Check resources before starting

File templates need sources and an output location. Knowledge base templates also need matching indexes and embedding models; table templates need a database and compatible fields. Follow the selected template’s preparation steps and start with a small sample.

Extraction templates combine inputs by default. Do not assume every file produces one row. When changing modes, also check downstream saving bindings. Each template page explains data flow, defaults, and result checks.

Scenarios

• Prepare text and images for RAG

• Extract resume information

• Generate legal fine-tuning data

Last updated on Sep 15, 2026