Content Preparation

Content preparation turns uploaded files into representations that search, agents, workflows, and downstream systems can use reliably. A project can convert documents to readable markdown, extract structured business data with Intelligent Document Processing (IDP), or combine both approaches.

The standard intake pipeline can:

1. assign a content type
2. convert the source to text or markdown
3. extract properties defined by the content type schema
4. verify and cite extracted values with IDP
5. render properties into canonical markdown when page conversion is unnecessary
6. generate text, property, and image embeddings

The pipeline is configured at the project level and refined for each content type. An invoice can use IDP to produce verified fields without converting every page, while a contract can retain a complete markdown representation and extract only selected terms.

Choose the right preparation path

Intelligent Document Processing

Intelligent Document Processing (IDP) is for documents whose structured fields drive downstream work: purchase orders, invoices, forms, applications, and similar records.

IDP extracts data against the content type schema and keeps source evidence for each value. It supports digital PDFs, scans, and handwritten content, can route difficult pages to a stronger model, and can run a review pass before declaring the result ready for downstream processing.

Use IDP when you need:

• structured properties rather than a page-by-page transcription
• citations that link values back to the source page
• separate digital, OCR, and AI verification signals
• an operational verdict such as good_to_go or needs_review
• events that can trigger a process, webhook, or human review

Markdown conversion

Markdown conversion transforms visually rich documents into structured markdown while preserving headings, reading order, tables, and figures. The resulting text becomes the document representation used by search, prompts, and agents.

Use markdown conversion when readers or models need the complete document narrative, especially for reports, contracts, policies, presentations, and other long-form material.

Property-based markdown

Some transactional documents do not need a full page conversion. After IDP extracts their properties, an intake rendering template can produce concise canonical markdown such as an order summary. This keeps search and agents useful without preserving page noise that has no downstream value.

Configure intake once, refine by type

Use Settings > Project > Intake to control automatic processing and set the project default policy. Then open a content type's Intake tab to override conversion, extraction, IDP, output, and embedding behavior for that type.

See Intake Configuration for the configuration hierarchy and common setups.