Overview

• Predictions are writes. Every create call uploads a file to Nanonets and consumes API quota, so they are classified write, not read. Files are sent as base64_data (PNG or JPEG); the binary multipart file field is not wrapped.
• async: true switches a create call to the documented async variant: the response comes back pending with a request_file_id, which you poll with predictions.getFileByFileId. Nanonets documents no done flag on that route: the file is processed once its pages are listed with their predicted_boxes, under unmoderated_images, or under moderated_images after the file is approved. Recommended above 3 pages; results normally finish within ~5 minutes.
• Quota and rate limits. 402 means the free API call quota is exhausted and is never retried. 429 is per model per 1-minute window on the predict calls; the plugin waits the documented 30 seconds between retries (Retry-After is honored when sent), up to three times.
• Paging is a day window. predictions.listFiles takes start_day_interval and current_batch_day as days since 1970-01-01 GMT (floor of unix seconds over 86400) and returns moderated and unmoderated images for that window; walk the windows forward yourself.
• files.updateFields replaces the whole field set of one page. Read the page first (predictions.getFileByPageId or predictions.listFiles), then send its id with the full moderated_boxes array: unchanged boxes stay, new boxes need is_new: true, and any field you omit is deleted. Copy day_since_epoch and hour_of_day from the same read: they are required, because Nanonets looks the page up by them and answers 400 without them. page and is_moderated are optional. The request is sent as text/plain because that is the documented working sample.
• Deleting is permanent. files.delete removes the file and its predictions for good. integrations.executeQuery runs arbitrary SQL against a connected store (documented examples include UPDATE), so it is also classified destructive.
• Training. models.uploadTrainingUrls adds images by URL (pass data to label them: one entry per image with its filename URL and the boxes in object), then models.train retrains; the model state walks 2→3→4→5 (usable). models.updateGuidelines changes extraction behavior of Instant Learning models via natural-language instructions.
• Webhooks. Nanonets can POST document events to URLs you configure in its UI (Workflow → Exports → Webhooks), but it signs nothing, so the plugin receives none.

• 17 typed API operations

​

Setup

npm install corsair @corsair-dev/nanonets better-sqlite3
npm install --save-dev @types/better-sqlite3

yarn add corsair @corsair-dev/nanonets better-sqlite3
yarn add --dev @types/better-sqlite3

pnpm install corsair @corsair-dev/nanonets better-sqlite3
pnpm install --save-dev @types/better-sqlite3

bun add corsair @corsair-dev/nanonets better-sqlite3
bun add --dev @types/better-sqlite3

import Database from 'better-sqlite3';
import { createCorsair } from 'corsair';
import { nanonets } from '@corsair-dev/nanonets';

export const corsair = createCorsair({
	plugins: [
		nanonets(),
	],
	database: new Database('corsair.db'),
	kek: process.env.CORSAIR_KEK!,
	multiTenancy: true,
	hub: {
		projectApiKey: process.env.CORSAIR_API_KEY!,
		signingSecret: process.env.CORSAIR_SIGNING_SECRET!,
	},
});

nanonets()

const { connectUrl } = await corsair.manage.connect.createLink({
	plugin: 'nanonets',
	tenantId: 'acme',
});
// redirect the user's browser to connectUrl

​

Example API calls

const tenant = corsair.withTenant('acme');
await tenant.nanonets.api.predictions.getFileByFileId({ model_id: 'd4d3beec-1029-497a-8d4c-8a3e2de6e7f2', request_file_id: '68d2b0f6-af9c-4c71-9d3e-2f3a4b5c6d7e' });

const tenant = corsair.withTenant('acme');
await tenant.nanonets.api.predictions.createFromUrls({
	model_id: 'd4d3beec-1029-497a-8d4c-8a3e2de6e7f2',
	urls: ['https://example.com/invoice.png'],
});

​

What’s next