EyePop.ai Node SDK

The EyePop.ai Node SDK provides JavaScript and TypeScript access to EyePop worker sessions for image and video inference.

Installation

Node

The SDK is tested with Node 20 LTS and newer.

npm install --save @eyepop.ai/eyepop

See the Node examples for runnable scripts.

Browser

Browser applications can use the bundled SDK directly:

<script src="https://cdn.jsdelivr.net/npm/@eyepop.ai/eyepop/dist/eyepop.min.js"></script>

API keys must not be used in browser code. Browser apps should connect with sessions created by your trusted backend.

React Native

React Native applications should use the React Native package:

npm install --save @eyepop.ai/react-native-eyepop

See the React Native example.

Authentication

The SDK supports these authentication patterns:

• API key authentication for trusted server-side code.
• Persistent worker session selection for provisioned deployments.
• Browser OAuth for dashboard users building local demos.

Server-Side API Key

Set EYEPOP_API_KEY in your server environment. API keys are secrets and must not be exposed in browser bundles, mobile app bundles, or public repositories.

export EYEPOP_API_KEY=<your_api_key>

Create transient sessions with a pop when the worker endpoint is constructed. This lets EyePop schedule the right compute before media is processed.

import { EyePop } from '@eyepop.ai/eyepop'
;(async () => {
    const endpoint = await EyePop.workerEndpoint({
        pop: {
            components: [
                {
                    type: 'inference',
                    ability: 'eyepop.person:latest',
                    categoryName: 'person',
                },
            ],
        },
    }).connect()
    try {
        const results = await endpoint.process({ source: { path: 'examples/example.jpg' } })
        for await (const result of results) {
            console.log(result)
        }
    } finally {
        await endpoint.disconnect()
    }
})()

For transient sessions, pass pop when creating the endpoint whenever possible. Use endpoint.changePop(pop) only when an already connected transient worker needs to switch Pops.

When sessionUuid is not provided, the SDK creates compute sessions with POST /v1/sessions?wait=true.

ModelLess CPU pops use the same constructor-pop path:

const endpoint = await EyePop.workerEndpoint({
    pop: {
        components: [
            {
                type: 'inference',
                ability: 'eyepop.localize-objects:latest',
                categoryName: 'objects',
                params: {
                    prompts: [{ prompt: 'person' }],
                },
            },
        ],
    },
}).connect()

You can also pass the key explicitly from server-side configuration:

const endpoint = await EyePop.workerEndpoint({
    apiKey: process.env.EYEPOP_API_KEY,
}).connect()

accessToken is accepted as the same bearer credential shape:

const endpoint = await EyePop.workerEndpoint({
    accessToken: process.env.EYEPOP_ACCESS_TOKEN,
}).connect()

The nested auth option is still supported but deprecated. Pass apiKey, accessToken, session, or oAuth2 at the top level for new code.

Persistent Sessions

If EyePop has provisioned a persistent worker session for your deployment, set EYEPOP_SESSION_UUID in the trusted server environment or pass sessionUuid explicitly.

export EYEPOP_API_KEY=<your_api_key>
export EYEPOP_SESSION_UUID=<your_session_uuid>

import { EyePop } from '@eyepop.ai/eyepop'
;(async () => {
    const endpoint = await EyePop.workerEndpoint({
        sessionUuid: process.env.EYEPOP_SESSION_UUID,
    }).connect()

    try {
        const results = await endpoint.process({ source: { path: 'sample.mp4' } })
        for await (const result of results) {
            console.log(result)
        }
    } finally {
        await endpoint.disconnect()
    }
})()

Persistent sessions are normally preconfigured. In that case, process media directly and do not call changePop() unless your deployment is meant to accept runtime Pop changes.

Persistent deployments are created through compute-api or EyePop tooling. For example, a trusted server can create a persistent deployment with an inline pop and then pass the returned session_uuid to the SDK:

curl -sS -X POST "https://compute.eyepop.ai/v1/deployments?wait=true" \
  -H "Authorization: Bearer $EYEPOP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "display_name": "node-modeless-deployment",
    "pop": {
      "components": [
        {
          "type": "inference",
          "ability": "eyepop.localize-objects:latest",
          "categoryName": "objects",
          "params": {
            "prompts": [{ "prompt": "person" }]
          }
        }
      ]
    }
  }'

The response includes session_uuid, session_status, status_url, and worker access metadata. Poll status_url until the deployment reaches pipeline_ok, then connect with sessionUuid.

Browser and Mobile Clients

Do not put EYEPOP_API_KEY in browser or mobile application code. Create EyePop sessions from a trusted backend, then pass only the session JSON to the client.

Browser OAuth for Local Dashboard Demos

Dashboard users can run local browser demos with the current browser session:

<script>
    document.addEventListener('DOMContentLoaded', async () => {
        const endpoint = await EyePop.workerEndpoint({ oAuth2: true }).connect()
        try {
            await endpoint.changePop({
                components: [
                    {
                        type: 'inference',
                        ability: 'eyepop.person:latest',
                        categoryName: 'person',
                    },
                ],
            })
        } finally {
            await endpoint.disconnect()
        }
    })
</script>

Processing Media

Local Files

const results = await endpoint.process({ source: { path: 'examples/example.jpg' } })
for await (const result of results) {
    console.log(result)
}

process() returns an AsyncIterable of predictions. Images normally produce one prediction. Videos and animated image containers produce one prediction per frame.

Readable Streams

Streams require an explicit MIME type:

import fs from 'node:fs'

const stream = fs.createReadStream('examples/example.jpg')
const results = await endpoint.process({
    source: { stream, mimeType: 'image/jpeg' },
})

Public URLs

Public HTTP(S), RTSP, and RTMP URLs can be processed without uploading the file from your application.

const results = await endpoint.process({
    source: { url: 'https://demo-eyepop-videos.s3.amazonaws.com/test1_vlog.mp4' },
})

Image Groups (multiple images, one result)

Send several images as a single source that the pop processes together as one inference unit — for example a multi-image VLM prompt. The group yields one prediction for the whole set, unlike batching below, where each image is an independent inference.

// Local files
const results = await endpoint.uploadGroup(['a.jpg', 'b.jpg', 'c.jpg'])
for await (const result of results) {
    console.log(result)
}

// In-memory streams (optional parallel MIME types)
import fs from 'node:fs'
import { Readable } from 'node:stream'
const stream1 = Readable.toWeb(fs.createReadStream('a.jpg'))
const stream2 = Readable.toWeb(fs.createReadStream('b.jpg'))
const results = await endpoint.uploadStreamGroup([stream1, stream2], ['image/jpeg', 'image/jpeg'])

// Remote URLs (the server fetches each)
const results = await endpoint.loadFromGroup(['https://example.com/a.jpg', 'https://example.com/b.jpg'])

Image order is preserved end-to-end. A group may contain up to 16 images (enforced server-side). The pop's ability must be multi-image-capable; a single-image ability handed a group returns an error.

Canceling Jobs

Queued and in-progress jobs can be cancelled from the result iterator:

const results = await endpoint.process({
    source: { url: 'https://demo-eyepop-videos.s3.amazonaws.com/test1_vlog.mp4' },
})

for await (const result of results) {
    console.log(result)
    if ((result.seconds ?? 0) >= 10) {
        results.cancel()
    }
}

Endpoint Options

By default, EyePop.workerEndpoint({ pop }).connect() starts a transient worker when needed and cancels queued jobs on that worker when it connects.

Disable worker startup:

const endpoint = await EyePop.workerEndpoint({ pop, autoStart: false }).connect()

Keep pending jobs:

const endpoint = await EyePop.workerEndpoint({ pop, stopJobs: false }).connect()

Visualization

Rendering helpers are provided by @eyepop.ai/eyepop-render-2d.

Data Endpoint Preview

EyePop.dataEndpoint() provides preview support for dataset management and model optimization workflows. This API is experimental and subject to change.

Composable Pops Preview

See Composable Pops for a preview of client-side Pop composition.