Nanonets

@Nanonets list all my OCR models and show me which ones have been trained in the last 30 days

Here are your Nanonets OCR models. You have 4 models total: 'Invoice Parser v2' (trained 12 days ago, 847 images), 'Receipt Extractor' (trained 8 days ago, 1,203 images), 'Contract Analyzer' (trained 22 days ago, 456 images), and 'Legacy Invoice v1' (trained 94 days ago, 312 images). The first three models fall within your 30-day window. Invoice Parser v2 and Receipt Extractor are your most active models based on recent training activity.

This example demonstrates the read-heavy 'Get all OCR models' tool. The response synthesizes metadata from multiple models. Note that you'll need a valid API key with read access to your Nanonets account. Model counts and training dates will vary by your actual account state.

@Nanonets create a new OCR model called 'Purchase Order Extractor' for processing vendor POs from our procurement inbox

I've created a new OCR model named 'Purchase Order Extractor' in your Nanonets account. The model ID is `mdl_8k3j9x2p`. This model is now initialized but not yet trained—you'll need to upload training images (typically 50-100 sample purchase orders with labeled fields) before it can process documents. Once trained, you can use this model to extract structured data from PO PDFs or images.

This showcases the 'Create OCR Model' tool, which initializes a new model but does not train it. After creation, you must upload labeled training images separately. The model won't be usable for predictions until training completes, which can take 15-45 minutes depending on dataset size.

@Nanonets pull all documents processed by my 'Invoice Parser v2' workflow in the last 7 days and summarize common extraction errors

I retrieved 143 documents from your 'Invoice Parser v2' workflow (last 7 days). Common extraction issues: 18 documents had low-confidence vendor name extraction (likely due to handwritten headers), 12 failed to parse line-item tables with merged cells, and 7 flagged mismatched totals between extracted subtotal and tax fields. 106 documents processed cleanly with >95% confidence across all fields. The handwritten-header issue appears most frequently on documents from vendor 'Acme Supplies'.

This combines 'List Workflow Documents' with AI reasoning to surface patterns. The MCP returns raw document metadata and confidence scores; the AI synthesizes trends. Useful for quality audits, but note that document counts and error types depend entirely on your workflow's recent activity and model accuracy.

When Nanonets handles recurring invoice layouts at scale

A 6-person finance team at a SaaS company processes 200+ vendor invoices monthly, each following one of twelve standard layouts. Nanonets wins here because you train one OCR model per vendor template, then route incoming PDFs through the matching workflow. The Create OCR Model and Get OCR Training Images tools let your team label 20-30 sample invoices per vendor in an afternoon, then the prediction endpoint extracts line items with 95%+ accuracy. If your invoices are one-off formats or handwritten, accuracy drops fast and you'll spend more time correcting than you save. For recurring B2B invoice formats, Nanonets cuts manual data entry from 40 hours to 4 hours per month.

Why Nanonets fits compliance-heavy document intake

A 3-person HR team at a 50-employee startup onboards 8-12 new hires per quarter, each submitting tax forms, ID scans, and signed offer letters. Nanonets is the right call when you need to extract structured fields (SSN, address, signature date) from semi-standardized government forms and route them into your HRIS. You create one workflow per document type using Get Workflows, then the List Workflow Documents tool shows which submissions passed validation. The threshold: if your forms vary wildly by state or country, you'll train 20+ models and the setup cost outweighs the benefit. For US-only onboarding with 3-5 repeating form types, Nanonets automates 80% of the data entry and flags edge cases for human review.

When Nanonets speeds up screenshot analysis in support queues

A 10-person customer support team receives 400 tickets weekly, 60% including screenshots of error messages or account dashboards. Nanonets helps when you need to extract error codes or account IDs from images before routing tickets to the right specialist. You train a single OCR model on your app's UI using Get OCR Training Images, then the prediction tools pull structured data from each screenshot attachment. The model works if your UI is consistent; if customers submit photos of monitors or heavily cropped images, accuracy falls below 70% and manual review takes longer than the automation saves. For SaaS products with stable UI and clear error states, Nanonets cuts ticket routing time from 15 minutes to 2 minutes per ticket.