POST /api/webhooks, POST /api/webhooks/:id/test
Document extraction webhooks tell your system when the work is done. Dynamite Docs sends an extraction.completed event to your endpoint, signs the raw body with HMAC-SHA256 and retries temporary failures. You can register up to five endpoints per account. No polling loop required.
When the completion event is sent
- Scripts that poll for new extractions burn requests and lag by the poll interval. Nobody wants their document pipeline to be as stale as the last cron run.
- An unauthenticated webhook is a spoofing hole. Anyone can POST a fake event, and a receiver that trusts it acts on data that never existed.
- Transient failures (a 429, a 5xx, a network blip) should be retried, not dropped. A lost event is a silent data gap.
- Webhook payloads that are not signed leave the receiver guessing whether the event is real.
How webhooks work
Dynamite Docs- 01Register an endpoint
- 02Get the secret
- 03Receive the event
- 04Verify and retry
Register an endpoint
Create a webhook with an HTTPS URL, up to five per account. Endpoints are validated as public URLs (SSRF-guarded) so the service never posts to internal addresses.
Get the secret
Your webhook secret is shown once at creation and used to verify signatures. You confirm the endpoint by its signature. Nothing is trusted on the URL alone.
Receive the event
When an extraction completes, we POST the extraction.completed payload with columns, rows, metadata, and confidence, signed in the X-Dynamite-Signature header.
Verify and retry
Verify the HMAC-SHA256 signature over the raw body before acting. Network errors, 429s, and 5xx retry with exponential backoff and jitter, up to three attempts.
Webhook payload and delivery
Dynamite Docs- Event
- extraction.completed
- Endpoints
- up to 5 per account
- Signature
- HMAC-SHA256 in X-Dynamite-Signature
The webhook contract is short enough to verify:
A realistic example
The payload your endpoint receives and how to verify it:
| Field | Example | Purpose |
|---|---|---|
| event | extraction.completed | Tells the receiver what happened |
| data.columns | ["Invoice #", "Date", "Total"] | The table headers |
| data.rows | [["INV-2041", "2026-06-03", "1240.00"]] | The extracted rows |
| data.confidence | 0.97 | Average confidence of the extraction |
| X-Dynamite-Signature | hmac-sha256 hex digest | Proof the event is genuine |
What webhooks do and do not
Does
- POST extraction.completed events to your HTTPS endpoints with an HMAC-SHA256 signature over the raw body.
- Retry transient failures (network, 429, 5xx) with exponential backoff and jitter, up to three attempts.
- Reject non-public/internal target URLs to keep delivery SSRF-safe.
Does not
- Does not send the full document bytes. Webhooks carry the extracted data, not the file.
- Does not retry forever: three attempts max, so a broken endpoint cannot be hammered.
- Does not deliver to endpoints that cannot verify the signature. Verification is the receiver’s job, and the secret is shown once.
Why webhooks complete the outbound story
Push beats poll. An event the moment extraction completes means your downstream systems act in real time, not on the last cron run.
HMAC signing makes delivery verifiable. The receiver can prove an event is genuine before trusting a single field.
Retries with backoff and an SSRF guard make the contract reliable and safe, so the webhook channel can be trusted for production traffic.
Setup and delivery questions
The extraction.completed payload: columns, rows, metadata, docType, and confidence. It carries the extracted data, not the source document bytes.
Compute HMAC-SHA256 over the raw request body using your webhook secret and compare it to the X-Dynamite-Signature header with a constant-time comparison.
Up to five per account. Each has its own secret and can be tested independently from the workspace.
Network errors, 429s, and 5xx retry with exponential backoff and jitter, up to three attempts with a per-attempt timeout.
Extraction guides
Related integrations
Extract document data using LLMs through a REST API
Extract document data using LLMs through the Dynamite Docs REST API. Submit files, retrieve structured rows, and manage scoped keys within plan limits.
/api/v1/extractions, /api/v1/infer, /api/tokens
Read the integrationPush reviewed AI-extracted tables to Google Sheets
Export reviewed AI tables from Dynamite Docs to Google Sheets. Connect once, push rows to a fresh spreadsheet tab, and keep document fields separate.
POST /api/sheets/export
Read the integrationExport reviewed AI extraction results to CSV, JSON, and Excel
Export reviewed AI data from Dynamite Docs to CSV, JSON, and Excel. Choose table downloads or Word and PDF exports from the reviewed document state.
Export API (?format=csv|json|xlsx)
Read the integrationDocument workflows it feeds
Prepare invoice data with AI for accounts payable teams
Prepare supplier invoice data with AI in Dynamite Docs for accounts payable. Review fields, line items, and tax before sending checked rows to your system.
Extract invoice line items for AI accounting
Extract invoice line items for AI accounting in Dynamite Docs. Check descriptions, quantities, prices and tax against the source before entry.
From the blog
Get told when extraction finishes.
Register an endpoint, verify the signature, and let extraction.completed events trigger the next step in your stack.