EyePop.ai
EyePop.ai is an SDK-first computer vision platform for developers and product teams. Compose detection, tracking, pose and OCR pipelines, train custom models without an ML team, and run the same pipeline on cloud, on-premise or edge hardware.
What is EyePop.ai?
EyePop.ai is a computer vision platform built by EyePop.ai, Inc., a San Diego company founded in 2023. Its core building block is the Ability: a pre-configured visual task made of a model, often a vision-language model such as Qwen3, a prompt, and a media sampling configuration that sets resolution, frame rate and analysis frequency. Abilities chain together into Pops, composable inference pipelines that execute in a single pass.
The primary interface is a software development kit, available for Python and Node/TypeScript, with a dedicated React SDK as well. A REST API makes the platform reachable from Java, C# and any other language, and every response comes back as JSON. Teams that prefer not to write code can work from the dashboard and the visual Workflow Designer, and a Claude Skill lets an AI coding agent build pipelines directly.
Out of the box, EyePop.ai covers object detection, 2D and 3D pose with facial mesh and hand tracking, OCR, persistent tracking, license plate reading and prompt-based scene understanding. Pre-trained models include dense captioning, zero-shot detection and vision-language models. When a stock model is not enough, Self-Service Training walks through five steps: define the target, upload or connect data, train, deploy, iterate. Auto-labelling and a built-in annotation interface cut the manual work; JPEG, PNG and MP4 are accepted, with frames extracted automatically from video, and the FAQ puts the starting point at roughly 200 annotated images. An Ability can then be evaluated automatically against a reference set, from the dashboard or the SDK.
Where a Pop runs is a decision separate from how it is built. The same definition executes unchanged on the EyePop cloud, on-premise inside a customer network, or at the edge on NVIDIA Jetson Orin boards, Qualcomm Snapdragon Dragonwing NPUs, GPU servers or CPU-only machines, with hybrid cloud offload over standard, cellular or Starlink links. Cloud consumption is metered in compute units that vary with resolution and analysis frequency.
The site addresses surveillance, marketplaces, broadcast, CDN, agriculture and livestock, construction, drones, insurance, roofing, PPE safety, traffic and agencies. Side products include Video Agent, which assembles multi-camera highlight reels automatically, and an SB-942 Watermark Compliance Detector. The platform took SIA Judges' Choice and Best Video Analytics 2026 at ISC West.
What it does
- Detect people, vehicles, animals and more than 50 object types, adding custom categories without retraining
- Train a custom vision model in a few hours from images or video
- Read text and license plates in real-world scenes, including on edge hardware
- Track object identity from frame to frame with persistent tracking
- Ask open questions about what a camera sees, through prompt-driven visual intelligence
- Turn surveillance video into structured, searchable events
- Check a video for California SB-942 compliance, covering watermarks and provenance data
When to use EyePop.ai / When not to
A quick filter to help you decide if EyePop.ai is the right fit.
When to use EyePop.ai
- Product teams and developers who want production computer vision without hiring machine learning engineers
- Video surveillance platform integrators enriching a VMS with structured, searchable events at ingestion
- Operations bound by data locality rules, where the on-premise runtime keeps video on the local network and sends only metadata upstream
- Engineering teams that need one pipeline to run unchanged on Jetson Orin, Snapdragon Dragonwing, GPU servers or CPU-only machines
- Teams starting from a small annotated dataset, since auto-labelling and roughly 200 labelled images are claimed to be enough for a first model
When not to use EyePop.ai
- Non-technical users looking for a point-and-click tool: the FAQ states the platform is built SDK-first for software developers and AI coding agents
- Small budgets: the lowest published cloud tier is USD 200 per month, and on-premise adds around USD 1,500 of hardware per box plus USD 5,000 of commissioning
- Organizations that require written GDPR guarantees: the site never mentions GDPR, publishes no DPA or subprocessor list, and names no EU representative
- Consumer or personal projects: the terms distinguish personal from business accounts, but the entire offer is aimed at organizations
- Anyone under 16, which the terms set as the minimum age
How to use EyePop.ai
A typical end-to-end flow, from setup to results.
- Create an account on the EyePop.ai dashboard, from the Sign in or Start Free link in the site navigation
- Collect your API key from the API Key section of the developer documentation
- Choose your entry point: compose a Pop from the Python or Node/TypeScript SDK, or work from the dashboard and the visual Workflow Designer
- For the fastest route, pick an existing Ability from the Abilities Hub and call it as is, with no training involved
- For a custom model, start the five-step Self-Service Training path by defining exactly what you want to detect
- Upload or connect your images, videos or streams, then annotate them in the built-in labelling interface or import an already annotated set through the Dataset SDK
- Launch training, which combines auto-labelling with sample prioritization
- Refine your wording with the Prompt Creation Agent, since vision-language models are sensitive to exact phrasing
- Deploy to the EyePop cloud, to the edge, or on-premise through the containerized On-Premise AI Application Runtime linked to your account
- Consume results downstream through API or SDK calls, JSON responses and webhooks, and retrain at any time by adding data
Pros & Cons
Pros
- One Pop definition runs unchanged from the cloud to Jetson, Snapdragon or a CPU-only machine, with no rewrite between targets
- Genuine on-premise mode: the video stream stays on the local network and only lightweight metadata is sent upstream
- Ownership of models, datasets and outputs is explicitly left with the customer, with no claim over trained models
- Fast custom training claimed: a few hours, around 200 annotated images, auto-labelling included
- Pre-trained models and a catalog of ready-made Abilities deliver value before any training is done
- Language-agnostic REST API alongside Python, Node and React SDKs, with JSON responses and webhooks for downstream integration
- Free three-month on-premise license for one internal lab box, so a deployment can be validated before hardware is committed
Cons
- The published pricing grid is unfinished: the feature lists on the pricing cards still show the placeholder text Feature text goes here
- No permanent free tier can be identified, even though the navigation offers a Start Free button and the cloud pricing grid lists free trial availability as none
- High entry point: USD 200 per month on the cloud side, and on-premise adds roughly USD 1,500 of hardware per box plus USD 5,000 of commissioning
- No GDPR mention, no DPA, no subprocessor list, no EU representative, and no published data hosting country or region
- No data retention period is published anywhere on the site
- The FAQ promise that data is never repurposed sits against terms allowing submissions outside Private User Input to improve EyePop.ai's models, and the ability to mark data as Private depends on an unstated tier
- Developer-first by design, with no mobile app and no browser extension; no SLA on Production and a custom SLA only on Enterprise; privacy policy unchanged since 31 January 2024
Pricing & Plans
EyePop.ai is sold on a subscription basis in US dollars, and no permanent free plan is offered: the cloud pricing grid explicitly lists free trial availability as none. The lowest published entry point is the Production cloud tier at USD 200 per month, which includes 4,000 compute units and 25 training iterations per month, with overage billed at USD 0.05 per compute unit. The Enterprise cloud tier starts at USD 800 per month, with two large dedicated servers and no overage charge. On-premise is licensed per box per month on a volume-degressive scale: USD 250 for 1 to 10 boxes, USD 200 for 10 to 100, USD 150 for 100 to 1,000, and by quotation beyond that. Hardware is purchased by the customer at an estimated USD 1,500 per box, and a one-off commissioning fee starts at USD 5,000, covering network configuration, integration with an existing surveillance platform and model tuning. The only documented free access is a software license for a single internal lab box for three months, which is an on-premise validation license rather than a cloud trial. Additional ML engineering hours, extra labelling and further training are quoted on a transparent, scoped-work basis.
- 25 training iterations per month
- 4
- 000 compute units
- general-purpose auto wake servers
- overage at USD 0.05 per unit
- data isolated from other tenants
- unlimited models
- production-ready infrastructure and team seats
- 25 training iterations per month
- two large dedicated always-on servers for low latency
- no overage charge
- redundant servers
- geo-distributed multi-tenancy
- guided onboarding
- tailored compliance support
- customizable SLA
- hardware purchased by the customer at around USD 1
- 500 per box
- commissioning from USD 5
- 000
- and a free three-month test license on one internal box
Data, GDPR & hosting
A consolidated view of how EyePop.ai handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the EyePop.ai website: no Article 27 EU representative, no data protection officer, no dedicated GDPR contact, and the only privacy address is info@eyepop.ai. No DPA is published or offered on request, and no subprocessor list exists. The privacy policy, last updated and effective 31 January 2024, is written for a US audience: no legal bases, no GDPR rights, no international transfer mechanism. The only consumer regulation named is the California Civil Code section 1789.3 notice in the terms. Governing law is the Federal Arbitration Act, California state law and US federal law; the terms state that foreign laws do not apply and place jurisdiction in San Diego County. The compliance claims made elsewhere on the site cover HIPAA and California SB-942, not GDPR. This is an absence of published information, not a finding of non-compliance.
Who owns the data?
EyePop.ai states that you keep full ownership and control of your models, datasets and outputs for as long as your account is active, and that it claims no rights over trained models; the Self-Service Training page describes deployment with 100% ownership retained. The terms require you to warrant that you hold every right in what you submit, and they sort incoming material into three levels: User Submission, covered by a very broad license granted to EyePop.ai, User Input, licensed only to operate the service, and Private User Input, the narrowest scope. Aggregated Statistics and derived ML Data belong solely to EyePop.ai, Inc.
Reuse rights
The terms allow User Submissions, excluding Private User Input, to be fed into or used to improve EyePop.ai's machine learning models, under a perpetual, irrevocable, worldwide, royalty-free license covering aggregated statistics and machine learning. Outside Private User Input, EyePop.ai also reserves the right to disclose a submission to a third party with no obligation of confidentiality. Whether you can designate an input as Private at all depends on the service tier you subscribe to, and the site never says which tier opens that option. This sits in open tension with the FAQ, which states that your data is your own and that EyePop.ai never repurposes, shares or analyzes it outside your specified use case. The privacy policy lists service delivery, support, analysis and improvement, communication, marketing, security, legal compliance, audit and operations as purposes, and treats aggregated, de-identified data as usable without restriction. On your side, you may reuse your own datasets, models and outputs freely while your account is active, with no permission to request. With the on-premise runtime, video never leaves your network and only lightweight metadata is transmitted.
Data retention & training
Hosting summary
No hosting country or region is published anywhere on the EyePop.ai website, and no data hosting jurisdiction is stated. On the cloud side, infrastructure is managed by EyePop.ai: the Production tier runs auto wake servers started on demand, while Enterprise runs two large dedicated servers kept permanently active and announces geo-distributed multi-tenancy. Isolation is addressed in general terms, with Production data described as isolated from other tenants and enterprise-grade security claimed for Enterprise and on-premise. The Physical AI page mentions end-to-end encryption and auditable data provenance. The privacy policy lists website hosting among the services entrusted to external providers, without naming any of them, and no subprocessor list is published. On-premise is the only configuration with a clearly defined location: processing happens on the customer's own hardware and network, video never leaves the local environment, and only lightweight metadata is sent upstream. One off-site technical observation, which is not a statement by the company: the resolved IP address belongs to a Cloudflare anycast network. Buyers with data residency obligations will need that answer in writing.
Things to keep in mind
Risks and trade-offs to weigh before adopting EyePop.ai.
- Outside Private User Input, the terms let submissions feed EyePop.ai's machine learning models and be disclosed to third parties with no obligation of confidentiality
- Whether an input can be designated as Private depends on the service tier subscribed to, and the site never says which tier opens that door
- Surveillance capability cuts both ways: person detection, persistent tracking, re-identification and plate reading make it easy to build monitoring that outruns what employees, customers or the public have agreed to, and model outputs treated as ground truth in safety or compliance decisions still deserve human review
- Mandatory individual arbitration with a class action waiver, exclusive California law, an explicit statement that foreign laws do not apply, and jurisdiction in San Diego County courts; opting out of arbitration is possible within 30 days of acceptance by emailing info@eyepop.ai
- The company may modify, suspend or discontinue the site at any time without notice, and gives no guarantee that a submission already sent can be edited or deleted
- HIPAA certification is claimed in the FAQ with no document, auditor or certifying body named
- Documentation gaps to raise directly with sales: placeholder text still on the pricing cards, no published retention period, no hosting country disclosed, and a privacy policy unchanged since 31 January 2024
Setup & Integrations
Technical difficulty
Moderate for a developer, heavy for anyone else. The lightest route is an account plus an existing Ability called through the API, no training involved. The standard route uses the Python or Node SDK and an API key to compose a Pop, deployed in under an hour according to the FAQ. A dashboard and visual Workflow Designer exist, and Self-Service Training claims no coding is required, which contradicts the FAQ describing an SDK-first platform, not a point-and-click tool. On-premise is another matter: hardware purchase, containerized runtime, surveillance system integration and commissioning from USD 5,000.
Deployment
Integrations
Behind EyePop.ai
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What is an Ability in EyePop.ai?
What is a Pop?
Do I need a dedicated machine learning team?
Which languages and SDKs are supported?
How much data do I need to train a custom model?
Can EyePop.ai run outside the cloud?
What is a compute unit?
Who owns the models I train?
Is my data used to train EyePop.ai's own models?
Can I test EyePop.ai before deploying it at a customer site?
Should you pick EyePop.ai?
EyePop.ai has a clear position: composability plus execution portability. Abilities are pre-configured visual tasks, Pops chain them into pipelines that run in a single pass, and the same Pop definition executes unchanged on the EyePop cloud, on a customer's own servers, or on Jetson Orin and Snapdragon Dragonwing hardware at the edge. That portability is the structuring advantage. Teams that cannot let video leave their network get an on-premise runtime where the stream stays local and only lightweight metadata travels, without maintaining two versions of the same pipeline.
The real audience is technical: engineering teams, video surveillance integrators, and software vendors that want computer vision without hiring a machine learning group. The FAQ is honest that no dedicated ML team is needed but that this is an SDK-first platform, not a point-and-click tool, which does not match the Self-Service Training page claiming no coding or technical expertise is required. Read the FAQ version.
Three reservations deserve attention before signing. Pricing documentation is unfinished, with placeholder text still sitting in the feature lists of the pricing cards. There is no GDPR framework of any kind: no mention of the regulation, no DPA, no subprocessor list, no EU representative, no published hosting country, no stated retention period. And the commercial promise that data is never repurposed sits against terms granting a perpetual license to train on everything outside Private User Input, with the right to mark data as Private depending on an unspecified service tier.
EyePop.ai is a young company, founded in 2023, funded with a USD 2.85 million round and recognized at ISC West 2026 with SIA Judges' Choice and Best Video Analytics. The technology is credible. The paperwork has not caught up with it.
- Choosing a selection results in a full page refresh.
- Opens in a new window.