LYNX ships late August 2026.Create a free account now — models, the browser demo and paid plans all open at launch.

Create free account →
Free account · 30-day demoAmerican MadeUpdated monthly

Production vision models.
Advanced SDK.

Production computer vision in one SDK. Detection, segmentation, depth, tracking, and 3D from a single camera — on six platforms.

Try it in your browser in 60 seconds. No install, no credit card.

6
Platforms · 1 SDK
60s
To your first inference
$0
To create an account
30d
Full-SDK trial
01 · Zero install

Try it before you install anything.

Create an account and run LYNX on your own images in the browser. Pick a model, set a confidence mode, drop an image. When you’re ready to go local, your API key and the SDK are one click away.

Create free account →30-day free demo with the models. Browser inference, the models catalog, API keys and your inference log.
02 · What ships today

Every capability below is in the box now.

No gating, no “talk to a rep” to find out what you get. What’s still being built lives on the roadmap, not here.

Perception

Detect, Segment, Depth, Pose

One model, one call. Boxes, pixel-accurate masks, per-pixel depth and keypoints come back together in the same result.

3D

3D from one camera

Real-world position and size of every detection. No stereo rig, no LiDAR, no second sensor.

Accuracy

Metric size filtering

LYNX knows how big things really are, and auto-rejects physically impossible detections before they reach your code.

Tracking

Tracking & temporal queries

Stable IDs across frames. Ask “how fast is track 47 moving” in one line.

Recognition

Scene text built in

Read labels, plates and signage from the same SDK. Vertical text and tiled high-res sweeps included.

Platform

Six platforms, offline

iOS, Android, Linux, Windows, macOS, Jetson. Runs fully offline after activation.

03 · And it actually wins

The model is smaller, faster, and more accurate.

This is what happens when you train on physics-accurate synthetic data with pixel-perfect annotations instead of crowd-labeled photographs. These numbers describe LYNX-trained models running on the SDK — bring-your-own-model results depend on the imported model.

Footprint · TensorRT engine
2.4MB

lynx-nano-w. Runs on a Jetson Nano. Fits in firmware. Ships in your APK without a fight.

LYNX
2.4 MB
YOLO11n
~5 MB
YOLOv8n
~5 MB
RT-DETR-L
~120 MB
Accuracy · ApplesM5 / Blaga 2025
44×

A 0.73M-parameter CNN matching a 32M-parameter transformer on its own benchmark. LYNX beats RT-DETR-L at 1/44 the size — and never saw a real image during training.

LYNX (synthetic)
0.778
RT-DETR-L
0.774
YOLO11n
0.563
YOLOv8n
0.561
04 · How it works

You run the models. We do everything else.

LYNX handles the entire model lifecycle. We acquire the training data, annotate it, train the models, and keep training them. You get updated models delivered automatically — typically every month — with no retraining, no data collection, and no pipeline work on your side. This is the LYNX-tier story. If you’re bringing your own model, the SDK tier gives you the runtime without this loop — see below.

After training, the updated model ships — typically every month
01

Model delivered

The current model lands on your hardware, signed and ready. No retraining, no pipeline work on your side.

02

Running in production

You run it on real footage. What it gets right and what it misses is the most valuable signal in the system.

03

Model training

Production feedback comes back to LYNX and goes straight into the training queue as a use case.

Your feedbackUse cases and images you send us
Audit feedbackRandom Audit scores, one click to escalate
New LYNX training dataSourced continuously, for every customer
Your model on day 400 is meaningfully better than your model on day 1 — and you did nothing to make that happen.
05 · Bring your own model

Want to use your existing models? No problem.

The SDK tier gives you the runtime without the Synetic model lifecycle. Import any ONNX, get the same six-platform C-first architecture, the same offline-by-design licensing, the same unlimited-deployment pricing. $12K/year, no per-device fees, no cloud round-trips.

Import any ONNX

YOLO, RT-DETR, or anything you can export. Calibrate on import against your own validation set, then ship it as a single signed .lnx artifact.

Ultralytics-compatible API

Your existing YOLO or RT-DETR code drops in without a rewrite — same call shapes, same CLI conventions, plus YOLO-format export back out.

What you keep, what you lose

You keep the C-first runtime, identical behavior on all six platforms, the offline license and unlimited-deployment terms. You lose monthly model updates and the model-dependent capabilities — 3D from monocular, dense depth, metric size gating, keypoints, behaviors, scene text. Those come with LYNX.

lynx_open_local("your_model.onnx")that is the whole migration.
06 · Built for production

Priced and licensed like you actually ship.

The things that decide whether a CV SDK survives contact with a real deployment.

Unlimited internal deployments

No per-device fees. Deploy to a million devices for the same price as one. OEM and redistribution are licensed separately.

Offline by design

The license validates locally. Your video never leaves your hardware — no cloud round-trip, no egress bill, no exposure.

American Made

Designed, trained and operated in the United States. No foreign-tech dependencies in the stack.

07 · Six platforms · one SDK

Install once. Deploy anywhere.

No PyTorch version pinning. No CUDA hell. No ONNX-conversion graveyard. C-first architecture means native bindings across six OS targets — same model behavior on every one.

P / 01
iOS
CoreML · Metal
◷ Beta
P / 02
Android
NNAPI · GPU delegate
◷ Beta
P / 03
Linux
CUDA · ONNX · CPU
◷ Beta
P / 04
Windows
CUDA · DirectML
◷ Beta
P / 05
macOS
Metal · Apple Silicon
◷ Beta
P / 06
Jetson
TensorRT · ARM64
◷ Beta
08 · Pricing

Annual subscription. Unlimited deployments.

Every plan starts with a 30-day free demo with the models — no card, no commitment. After that: no per-device fees, no per-call billing, and public pricing so you never need a sales call to find out what it costs.

Free account
$0
Instant · no credit card
  • 30-day free demo with the models
  • Run inference in the browser
  • Models catalog + API keys
  • Inference log and feedback loop
Create free account
SDK
$12K/year
Bring your own model
  • Full SDK runtime · all 6 platforms
  • Import any ONNX, calibrate on import
  • 100 devices (+overage) · unlimited dev seats
  • Ultralytics-compatible API and CLI
  • Email support
Get started
Recommended
LYNX
$24K/year
$20K with single-payment discount
  • Everything in SDK
  • Synetic-trained models · monthly updates
  • LYNX street category, new models as released
  • 3D from monocular depth, dense depth, metric size gate
  • Keypoints + LOD, behaviors, scene text
  • Priority email support
Get started
Enterprise
Custom
Volume pricing available
  • Everything in LYNX
  • Custom-trained models for your domain
  • Unlimited devices per site
  • SLA + named contact
  • Custom training pipeline
  • Priority on monthly update queue
Talk to sales
30-day trial — full SDK, demo model, no credit card. It starts when you activate, not when you sign up, so creating an account early costs you nothing.
The full comparison

Every paid tier gets the full runtime. The tier decides what models you run.

Every paid tier gets the full runtime and all 28 model-agnostic features — listed below, no asterisks. The table covers only what actually changes between them: device count, support level, and the capabilities that require synthetic-data-trained models. The free account is the 30-day demo, not a tier — see pricing above.

In every paid tier
Plan
  • Unlimited dev seats
Models
  • Import your own (ONNX)
  • Calibrate on import (your val set)
Runtime & deployment
  • Six platforms, one SDK
  • Language bindings
  • DeepStream integration
  • Single-artifact packaging (.lnx / .lnxp)
  • TRT engine management
  • Offline validation, no cloud in hot path
  • Prewarm / provider override / precision
Video & multi-stream
  • StreamManager (N cameras, one session)
  • RTSP auto-reconnect
  • Crash-safe video writer
  • Multi-object tracking
  • Track value operators
Analytics
  • LYNXZone / LYNXLine primitives
  • Zone dwell tracking
  • Built-in HTTP server
Control & tuning
  • PR curve modes
  • Per-class conf thresholds & callbacks
  • Autotune diagnostic
  • RGB/BGR, resize, normalize auto-detect
Migration
  • Ultralytics-compatible API
  • Ultralytics-style CLI
  • YOLO-format export
Feedback loop
  • Low-confidence flagging
  • submit_feedback
  • Hard telemetry opt-out
Feature
SDK
LYNX
Enterprise
Plan
Free demo
30 days
30 days
30 days
Price
$12k/yr
$24k/yr
Quoted
Devices
100 (+overage)
100 (+overage)
Unlimited / site
Support
Email
Email, priority
SLA, named contact
Models
LYNX street category
New street models as released
Monthly model updates
Custom-trained models
Requires LYNX-trained models
Detection
your model
3D from monocular depth
Dense depth
Metric size plausibility gate
Object-to-object distance
Four heads, one pass
Per-detection lazy head realization
Keypoints + LOD scaling
Behaviors
Scene text
Services
Fleet distribution
Add-on
Managed conversion
Quoted
Quoted
Synetic data SKUs
Quoted
Quoted
Included allowance
09 · Get started

Three steps to running code.

Nothing to install until step three, and nothing to pay until you’ve already seen it work on your own images.

→ STEP 01
Create an account.
Instant, and the 30-day free demo starts here. No credit card, no sales call. You land in the dashboard with an API key already minted.
▣ STEP 02
Run it in the browser.
Pick a model, drop one of your own images, and look at the detections. Sixty seconds, zero install, on your data rather than our demo reel.
⌘ STEP 03
Go local.
pip install lynxwith your key when you’re ready. Same model, same results, now on your hardware and fully offline.

Two doors from here.
Both start with an account.

Start free — the models, in your browser, no card, 30 days. Then buy the SDK at $12K/yr or LYNX at $24K/yr, public pricing, no sales call.