Same device. Same ID.
Open-source device identification and fraud signals, self-hosted. Recognise returning visitors across private windows, cleared cookies and other browsers, even when they stay anonymous.
Or open inCursorClaudeinstall.md
Hello, visitor ID
- Private window
- VPN
- Bot
- Spoofing
Identifying
Real demo, run against a Trace test server. Accuracy figures on this page are synthetic. What the demo stores
Works with your stack
- JavaScript
- TypeScript
- React
- Next.js
- Vue
- Svelte
- Angular
- Solid
- React Native
- Flutter
- Kotlin
- Swift
- Node.js
- Python
- PHP
- Java
- Go
- WordPress
- Cloudflare
Identification
The ID survives.
Cookies come and go; the device stays recognisable. Pick a situation and see what Trace keeps, and how sure it is.
How matching worksyourapp.com/signup- Cookiegone
- Storagegone
- Private modeoff
- Device traitsmatched
A visitor ID is issued and a first-party cookie set.
Fraud signals
A score you can take apart.
Switch signals on and off. The score is a readable sum of weighted signals, every reason is itemised, and your backend turns it into a decision.
All 26 signalsStart from
Suspect score
Block- Bot / automation
+60 - Virtual machine
+15 - Datacenter IP
+15
Default weights from scoring.default.json. Tune them, or learn them from your own outcomes.
Suspect score 90 of 100. Decision: Block. 3 signals on.
Use cases
One event. 24 ways to use it.
The same verified result answers sign-up, login, checkout and moderation questions. Pick one to see the signals it leans on.
See all 24 use casesLogins to one account
| Device | Network | Decision |
|---|---|---|
MacBook, SafariKnown device for this account | Home, Cairo | Allow |
iPhone, SafariKnown device | Mobile, Cairo | Allow |
Windows, ChromeNew device, hosting IP: ask for a code | Datacenter, NL | Review |
Headless Chromebot and velocity | Proxy, US | Block |
The whole stack
Everything a fraud check needs, in one service.
Identification, network and automation signals, scores with reasons, and the plumbing to act on them. One call from the browser, one verified read on your server.
Device identification
One visitor ID through private windows, cleared storage and updates.
IP intelligence
VPN exits, Tor, hosting networks and your own blocklist.
Bot detection
Webdriver flags, headless browsers, robotic input and TLS mismatches.
Scores with reasons
A 0 to 100 score you can read: what fired and how many points.
Anti-detect browsers
The page, a worker and an iframe disagree about the device.
Device graph
Every account on a device, every device behind an account.
Signed webhooks
Events pushed to your backend, signed with HMAC-SHA256 and retried.
Runs on your servers
Postgres, Redis, the API and the dashboard. Data never leaves.
Values in these cards are examples.
Install with AI
One prompt. Your agent does the integration.
The prompt points your coding agent at a guide written for agents: which SDK fits your stack, where to identify, how to verify on the server and what to do when Trace is down. You review the diff.
claude in your app's repository and paste.Add Trace device identification to this app. Read https://trace.dev/install.md and follow it step by step. Ask me for TRACE_ENDPOINT, TRACE_PUBLIC_KEY and TRACE_SECRET_KEY if they are not already in the environment, and never commit real keys. Show me the diff when you are done.
- Read install.md
- Found Next.js (app router) with API routes
- npm install @trace-id/sdk-react @trace-id/sdk-node
- app/providers.tsx+9
- app/signup/form.tsx+12 −2
- app/api/signup/route.ts+21
- const e = await verifyEvent(eventId, { secretKey, endpoint });
- if (e.suspectScore >= 60) return deny(e.eventId);
- .env.example+3
- npm test14 passed
- Signup now sends an eventId; the API verifies it before creating the account.
Integrate by hand
Identify in the browser. Decide on your server.
The browser only hands over an event ID; your backend reads the verified result with a secret key.
Integration guideimport { createClient } from '@trace-id/sdk-js';
const trace = createClient({
apiKey: 'pk_...',
endpoint: 'https://id.yourapp.com',
});
// On signup: identify, then send the event ID with the form.
const { eventId } = await trace.get({ linkedId: user.id });
await fetch('/api/signup', {
method: 'POST',
body: JSON.stringify({ ...form, eventId }),
});{ "eventId": "0193a3f2-7c1e-7b4a-9d2f-5e8c1a6b3f40", "visitorId": "Q7mZ2xKp9LwR4tVb8NcE", "deviceId": "Q7mZ2xKp9LwR4tVb8NcE", "matchKind": "new", "confidence": 1, "suspectScore": 90, "reasons": [ { "signal": "bot", "contribution": 60 }, { "signal": "datacenterIp", "contribution": 15 }, { "signal": "virtualMachine", "contribution": 15 } ], "signals": { "bot": { "detected": true }, "vpn": { "detected": false }, "incognito": { "detected": false } }, "ip": { "country": "US", "asn": 64501 }, "linkedId": "user_48213"}Accuracy
Measured on every change.
Matching changes must not regress an accuracy run before they ship. The figures below come from that run, on simulated devices with private windows, cleared storage, updates and travel.
How it is measuredSynthetic data, not real traffic. 1,000 simulated devices. Real-world numbers will differ and are likely lower. Method
Privacy
What it reads. What it never touches.
Collectors do not prompt for permissions, make network requests or read devices, files or location. High-entropy readings are hashed in the browser, and data and IPs stay on your servers.
Privacy detailsReads
- Screen and GPU
- Fonts and audio
- Time zone and languages
- Pointer rhythm, never text
Never touches
- Permission prompts
- Location
- Files
- Camera and microphone
Hashed in the browser: canvas a91f3c… audio 07be2d… fonts c4e810…. Details
Clear your cookies, then open the demo.
Pre-release, open source, no customers yet. Run it, try to fool it, tell us where it breaks.