Description
n8n-nodes-deepread
This is an n8n community node for DeepRead — AI-native document processing with two engines: Extract (one OCR pass) and Deep Extract (two OCR passes, an LLM judge and a second read that checks each extracted field value).
n8n is a fair-code licensed workflow automation platform.
Installation
Operations
Credentials
Compatibility
Usage
Resources
Installation
Follow the installation guide in the n8n community nodes documentation.
Quick install in n8n:
1. Go to Settings → Community Nodes
2. Click Install
3. Enter n8n-nodes-deepread
4. Click Install
Operations
The DeepRead node supports four operations, all powered by the same DeepRead API:
OCR Extract
Extract text from any PDF or image. Returns the raw extracted text, plus a preview URL for the document when the Preview option is on.
Accepted uploads depend on the plan: PDF, PNG and JPEG on Free; TIFF, WebP, BMP, GIF, DOCX and TXT from Standard; office, spreadsheet and HTML formats on Enterprise. Anything that is not a PDF, PNG or JPEG is converted to PDF before the page count, and the page count is what is charged.
Structured Extraction
Extract typed JSON fields from documents using a custom JSON Schema. The node outputs the full job response. Extracted fields live in the extraction.fields array ($json.extraction.fields), where each field is an object:
key— the field name from your schema (e.g.invoice_number)value— the extracted value (scalar, object, or array)needs_review—trueif the field needs human reviewreviewreason— explanation when flagged (present only whenneedsreviewistrue)location—{page, bounding_box}, where the value was found (when the Bounding Boxes option is on)- Plan Default — omits the field: Free and Standard plans run Extract, Enterprise plans run Deep Extract
- Extract — one OCR pass: Markdown, layout and bounding boxes, plus your fields when you send a schema
- Deep Extract — two OCR passes, an LLM judge and a second read that checks each extracted field value
- Bounding Boxes — locate layout blocks and extracted values on the page (sent as
include_markers): agroundingarray plus alocationon each field. Works on both engines and every plan - Idempotency Key — a retry with the same key returns the job the first request created instead of a second one; a different request with the same key is refused with
409 - Incognito (Enterprise) — the document is encrypted with a single-use key and crypto-shredded the moment the job finishes. No preview link; cannot be combined with Searchable PDF or Reviewer Email
- Per Page — a
pagesarray with the Markdown, layout and fields of each page - Preview (default off) — page images and a public preview link. Anyone with the link can open the document, so it exists only when you turn this on. Add Bounding Boxes for click-to-highlight in the preview
- Retention (Days) (Enterprise, 1–365) — the document, preview artifacts and extracted results are deleted that many days after submission
- Reviewer Email, Review Due (Hours), Review Expiry (Days) — have a person review the document in DeepRead; see Human review
- Webhook URL (Standard and up) — an HTTPS URL that receives the job result and the review events; see Webhooks
- Reviewer Email — one email, or a team as several emails separated by commas (up to 10). Any of them can review it; the first to complete the review completes it. Each reviewer is emailed a link when the document is ready and signs in with a code sent to them
- Review Due (Hours) — the review is due this many hours after submission (1–8760). When it passes,
review.overdueis sent and the review stays open - Review Expiry (Days) — an unfinished review closes this many days after submission (1–365, default 30, never later than Retention).
review.expiredis sent with the unreviewed data filledformurl— URL to the filled PDFfields_detected— total fields foundfields_filled— fields successfully filledfieldshilflagged— fields needing review- Names, SSNs, credit cards, emails, phones, addresses
- IPs, DOB, passport numbers, driver’s licenses
- Bank accounts, IBANs, URLs, medical record numbers
- Tested with n8n 1.0+
- Works in self-hosted and cloud n8n
- Compatible with all standard n8n agent integrations (use as a tool in AI Agent nodes)
Use this for invoices, receipts, contracts, medical records, insurance claims — any document where you need specific fields.
Engine and options (OCR Extract and Structured Extraction)
Engine picks how the document is processed and is sent as the API’s pipeline field:
Every job reports its product (extract or deep-extract) and the pipeline you asked for. The older names fast and standard still work as aliases for extract and deep-extract.
Searchable PDF also returns a searchable/selectable PDF as a binary output (searchablePdf). It is a Deep Extract add-on on Enterprise plans; the Extract engine does not support it.
Options (add the ones you need):
A feature the plan does not include is refused with 402 and a body naming the plan that has it.
Human review (Reviewer Email)
Name who reviews and DeepRead runs the review: no review screen, queue or audit trail to build.
The node returns as soon as the job completes, with review.status = pending, review.reviewers, review.url, review.dueat and review.expiresat. The reviewed data arrives later at your Webhook URL as review.completed (and review.updated for every later fix), with the same body as the job plus review.completedby, review.version and review.changes[] ({key, previousvalue, value, changedat, changedby, version}, one per corrected value). Reviewer Email turns Preview on and cannot be combined with Incognito.
Form Fill
Fill any PDF form (including scanned, non-editable forms) with your data. AI visually detects fields, semantically maps your JSON data, fills the form, and quality-checks the result. Returns the filled PDF as a binary attachment. Enterprise plans.
Returns:
PII Redact
Detect and redact 14 types of PII with irreversible black bars, built for HIPAA/GDPR redaction workflows:
Returns the redacted PDF as a binary attachment plus a detection report by PII type. Enterprise plans.
Supports 5 languages: English, Chinese, Spanish, Hindi, Arabic.
Credentials
You need a DeepRead API key:
1. Sign up at deepread.tech/dashboard (free, no credit card) and accept the Terms when asked
2. Get your API key
3. In n8n, create a new DeepRead API credential
4. Paste your API key (starts with sklive...)
5. Click Test to verify
Free: 2,000 pages a month, resetting on the day you signed up; up to 50 pages per document
Standard: prepaid credits — Extract $15, Deep Extract $35 per 1,000 pages; up to 1,000 pages per document; webhooks and blueprints
Enterprise: custom pricing; up to 2,000 pages per document; adds PII Redact, Form Fill, Searchable PDF, Incognito, Retention and bring-your-own LLM key
An API key can also carry a monthly spend limit (set in the dashboard). A job that would pass it is refused with 402 and keylimitreached, naming when the limit resets.
Compatibility
Usage
Example: Auto-process invoices from email
[Gmail Trigger] → [DeepRead: Structured Extraction] → [Google Sheets]
Schema: vendor, total, due_date
Example: Redact PII before LLM processing
[File Upload] → [DeepRead: PII Redact] → [OpenAI: Analyze]
Example: Auto-fill onboarding forms
[Webhook] → [DeepRead: Form Fill] → [Email: Send filled PDF]
Form fields: full_name, dob, ssn, address
Example: Human review before posting
[Dropbox Trigger] → [DeepRead: Structured Extraction] → (job returned, review pending)
Reviewer Email: [email protected]
Webhook URL: the Webhook node's URL below[Webhook: review.completed] → [IF: $json.review.status == "completed"] → [ERP: Post invoice]
The node handles all async polling automatically — submit a document, get the result back. No need to manage job IDs or polling logic. A poll that hits the plan’s polling limit (429) or a transient 503 waits for Retry-After and tries again until the Timeout.
Webhooks
Webhook URL (Standard plans and up) delivers one POST when the job completes or fails, with the same body as the job response, and for Reviewer Email jobs the review events review.completed, review.updated, review.overdue, review.cancelled and review.expired. The event name is in the body as event and in the X-DeepRead-Event header. Point it at an n8n Webhook node to continue the workflow when the review is done.
Every delivery is signed with X-DeepRead-Signature (t=, an HMAC-SHA256 over GET /dashboard/v1/webhooks/secret.
Resources
Version history
0.5.0
review.completedinclude_markers), Per PageRetry-After on 429 and transient 5xx instead of failing the item0.4.1
product0.4.0
0.3.0
pipeline field0.1.2
{code, message} objects
