Loading Precidoc…
Loading Precidoc…
Precidoc is an early access document preparation product by RAinc. The current workflow runs in your browser: real text parsing, explainable rules, editing, and reviewed exports.
| Limit | Public demo | Free |
|---|---|---|
| Active documents | 1 | 10 |
| MB per file | 5 | 15 |
| Pages per PDF | 20 | 200 |
| Characters per file | 100000 | 750000 |
| Files per batch | 1 | 5 |
| Seconds per file | 30 | 30 |
Illustrative before/after examples. Corrections and metadata are supplied by the reviewer, not an LLM.
Restore the latest backup. Where did this instruction come from?
Title: Database recovery Section: Recovery procedure 1. Verify the backup. 2. Restore the approved snapshot. Source: section_03 · edited by reviewer
Review the prerequisites with the procedure, then export referenced chunks for your retrieval pipeline. Precidoc does not create embeddings or measure retrieval accuracy.
Try this step in the sampleImport a digital PDF, DOCX, UTF-8 TXT, or Markdown file. PDFs keep page references; DOCX and Markdown keep section references. DOCX tables become Markdown tables.
The public sample is parsed from a predefined Markdown runbook. You may replace it with one local file within the demo limits. The homepage walkthrough starts from pre-parsed sample text and is labelled accordingly.
Checks identify missing metadata, empty sections, unreadable replacement characters, exact repeated sections, and extraction limitations. Findings link to the relevant source where a page or section is available. Compare original extracted text with your edits.
The readiness score is a weighted preparation checklist. It does not assess factual accuracy, detect all sensitive data, predict AI answer quality, or benchmark retrieval.
Inspect chunks and their references, confirm your review, and download Markdown. Registered accounts also get JSON, document metadata controls, chunk sizes, and exact-section deduplication. Editing invalidates previous approval.
No OCR for scans, LLM review, semantic deduplication, embeddings, vector-database publishing, cloud document storage, team sharing, or persistent document history. Multi-column PDF reading order and PDF tables need manual comparison.
Optional Claude/Anthropic assistance is planned. It will require its own processing disclosure and opt-in before any document content is sent to a model provider. It is not part of the current Free workspace.