Document AI.
Turn documents into structured data — forms, tables, invoices, contracts — grounded in the model, not brittle templates.
Extraction, classification and validation that turn PDFs and scans into structured data — layout-aware and grounded in the model, so it reads a document it's never seen instead of breaking on a new template.
- Category
- Applied AI
- Deployment
- Managed → air-gapped
- Governance
- IAM · encryption · audit
Template-based extraction breaks on the document it's never seen.
The traditional way to turn documents into data is to build a template per layout — and it works right up until a vendor changes their invoice, a new form arrives, or a scan is skewed. Then someone rebuilds the template. Aether's document AI is layout-aware and grounded in the model, so it reads a document by understanding it, not by matching a fixed shape — a new invoice format is just another document, not a project.
Extract, classify and validate — then route what it isn't sure about.
It pulls key-value pairs, tables and line items with positional awareness, classifies and routes documents by type, and validates the result — checking that totals reconcile, dates are sane and references resolve. Crucially, low-confidence fields are sent for human review rather than guessed, so the output you trust is genuinely trustworthy and the things that need a person actually reach one.
On the platform, so extraction flows straight into the work.
Because it runs where your data and the model live, an extracted invoice doesn't get exported to a separate system to be useful — it lands in the warehouse, triggers a workflow, or feeds an agent, under one identity and one audit trail. And it deploys in your boundary, which matters when the documents are contracts, KYC packets or medical records.
Extract structured fields from any invoice layout, validate the totals, and route only what's uncertain to a person.
import { aether } from "@aether/sdk";
const doc = await aether.documents.extract({
file: invoicePdf, // any layout, no template
fields: ["vendor", "total", "due_date", "line_items[]"],
validate: "line_items.sum == total",
});
if (doc.confidence < 0.9) routeToReview(doc); // not guessed
else postToLedger(doc.fields);No template to build or maintain: the model reads the invoice whatever its layout, checks that the line items sum to the total, and sends the genuinely ambiguous ones to a human instead of silently guessing.
Three steps to running.
PDF, scan or photo — no template to build first.
Layout-aware key-values, tables and line items, sorted by type.
Totals and references checked; low-confidence fields routed to a human.
Document AI, in full.
Key-value pairs, tables and line items, positionally aware.
Sort and route documents by type automatically.
Check totals, dates and references; flag what doesn't reconcile.
Low-confidence fields routed for review, not guessed.
Provision it in a few lines.
Every service is reachable from the same SDK, CLI and infrastructure-as-code — one identity, one bill, one audit trail across the whole catalog.
# Provision document ai on Aether Cloud
aether applied-ai create \
--service document-ai \
--name app \
--region us-1 \
--deploy managed # or vpc | on-prem | air-gappedAt a glance.
- Inputs
- PDF · scan · photo
- Extraction
- KV · tables · line items
- Templates
- None required
- Validation
- Built-in reconciliation
- Review
- Human-in-the-loop
Built for real work.
Invoice & receipt processing
Contract analysis
Onboarding & KYC
On one model, not stitched together.
The usual stack runs document ai in one product, the model in another and the data in a third — and the seams between them are the cost. Aether Cloud runs it on the same platform that serves the model, governs your identity and deploys into your boundary, with the rest of the catalog one hop away.
No stitching a vector DB to one place, a warehouse to another and a model to a third — document ai sits next to the rest of the catalog, one identity, one bill.
The provider that runs Aether runs your document ai — so the data and the model never leave the same governed boundary to talk to each other.
Need a capability that isn’t here yet? The model writes and deploys it into the same boundary — the catalog is a starting point, not a ceiling.
Good to know.
Layout-aware extraction and classification — grounded in the model, so it reads a document it's never seen instead of breaking on a new template.
No — it generalizes across layouts; few-shot specialization is there when you want higher accuracy on a specific form.
Low-confidence fields are routed for human review, not silently guessed.
Your boundary, your choice.
Pairs well with.
Detect, classify and read images and video through one API — grounded in the multimodal model, not a fixed label set.
Speech-to-text and text-to-speech in one service — transcription, diarization and natural voices, on the model that reasons over what's said.
Real-time and batch translation across 100+ languages, with glossaries and the context the model carries — not phrase-book substitution.
Time-series forecasting as a service — demand, capacity, load — with uncertainty intervals and what-if scenarios.
Real-time recommendations and next-best-action from your interaction data, served at low latency.
Run Document AI on Aether Cloud.
Turn documents into structured data — forms, tables, invoices, contracts — grounded in the model, not brittle templates. Deployable managed, in your VPC, on-prem or fully air-gapped — talk to us about the configuration your workloads and your boundary require.