PDF Intake Check — Page Audit & Text Chunks

Check what text your PDF exposes before sending it into an AI search or summarization workflow. Returns a page-by-page audit and overlapping text chunks with physical PDF page references.

Quick start

1. Encode one PDF as base64, without a data-URL prefix.
2. Paste it into pdfBase64 and run.
3. Open the default dataset for page results. In the default key-value store, open REPORT for the complete audit and CHUNKS for text chunks.
4. When reviewRequired is true, inspect flagged pages before continuing your pipeline.

Outputs

Dataset rows contain page, status, characters, and warnings. Statuses: text_extracted, no_extractable_text, sparse_text, extraction_error.

In a verified three-page cloud sample, readable text returned 83 characters, a blank page returned zero characters with no_extractable_text, and a page containing Short returned six characters with sparse_text. Blank and image-only pages can both return no text; inspect the original before deciding whether OCR is needed.

REPORT includes SHA-256, page/character counts, reviewRequired, flaggedPages, pages, chunks, and limitations. CHUNKS contains the same chunk list separately. Chunks include identifier, source, physical page number, start/end character offsets, and text. Offsets start at zero with an exclusive end and refer to extracted text. Chunks use 1,800 characters with 150-character overlap and never cross pages. The current adapter labels the source document.pdf; use SHA-256 to distinguish documents. Identical bytes yield stable identifiers with the same parser version.

Limits

• One PDF per run; at most 20 MB decoded, 300 physical pages, and 2 million extracted characters.
• Sparse means fewer than 40 non-whitespace characters.
• Encrypted PDFs and invalid inputs are rejected; the run is marked failed without a report.
• Parser timeout: 45 seconds. Use at least a 60-second overall timeout; tested memory: 512 MB.
• No OCR, table reconstruction, semantic validation, malware scan, reading-order guarantee, or printed-page-label detection.
• No review flag does not certify complete or correct extraction.

Data handling

Apify stores run inputs and outputs according to its storage settings. No zero-retention promise. Use synthetic or public documents for this pilot. This tool does not fetch arbitrary URLs or forward text to an external AI model.

Experimental pilot

Price: $0.02 per completed PDF audit, with platform usage included. One document is one event, regardless of its page count within the stated limits. A completed audit that flags blank or sparse pages is still a billable result. Rejected inputs and parser timeouts do not trigger the document event. The synthetic prefilled example is also billable when run. No separate start or per-page fee. Small cloud cases have been checked; performance on every PDF is not assured.

For reproducible issues, use Apify issues with a synthetic or public sample. Do not post confidential documents or credentials.