Roboflow plugin

• api.roboflow.com for management: workspaces, projects, versions, models and image upload.
• serverless.roboflow.com for inference. It is the Inference server’s interface, not the Platform API’s, with image URLs in the image query parameter and base64 images in the raw form body. Other options are query parameters.

• account.get returns 200 whether or not your key works. With no key it still answers a welcome message and instructions, so a successful status proves nothing. Only workspace appears for a valid key, and that is what the plugin’s authenticated field reports. Reading the 200 as a success is the easiest mistake to make here.
• An inference model ID is project/version. Take it from models.list’s url, which is exactly what infer expects. A model not tied to a version is workspace/model and splits the same way; the plugin rejects an ID with no separator rather than sending a request that cannot resolve.
• infer is billed per image. It is marked write for that reason and is never retried. The prediction payload is the Inference server’s and differs by task: a detection model returns boxes in predictions, single-label classification returns class scores in an array, multi-label classification returns a class-keyed object, and semantic segmentation returns a mask object. Instance segmentation adds points to each prediction.
• A duplicate image upload is a success. images.upload answers duplicate: true with the existing image’s id and adds nothing. That is not a failure, and treating it as one would drop a valid image ID.
• images.upload accepts a project slug or a workspace/project ID. Both use /dataset/{project}/upload with the project slug. The API key selects the workspace.
• annotation is a shared label group. Projects in the same group share classes, so reusing one is how a second project starts from an existing label set. anomaly-detection additionally needs the feature turned on and a group no other project type uses.
• license is required on a public plan. Private is what keeps a project private; any other value publishes it on Universe, and a public plan cannot create a private project at all.
• versions come back newest first, and version.generating tells you whether the version is still being built — a progress under 1 means the image count is not final.

• 10 typed API operations

​

Setup

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

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

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

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

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

export const corsair = createCorsair({
	plugins: [
		roboflow(),
	],
	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!,
	},
});

roboflow()

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

​

Example API calls

const tenant = corsair.withTenant('acme');
await tenant.roboflow.api.models.list({ workspace: 'roboflow', project: 'chess-sample-4ckfl' });

const tenant = corsair.withTenant('acme');
await tenant.roboflow.api.infer({ modelId: 'chess-sample-4ckfl/3', image: 'https://example.com/board.jpg', confidence: 0.5 });

​

What’s next