See every session.
Hear what your customers can't say.
DOM, network, console, marketing tags, journey signals — captured live from every visitor, PII shielded at the source. AI turns the noise into ranked insight for every team.
Prefer to read? Everything's on the page below — scroll on.
Feature
Session Replay That Actually Helps
Pixel-perfect DOM reconstruction, AI summaries on every recording, and a frustration heatmap that points to the exact moment users gave up. No video files. No 1990s-style mouse trails.
Why teams pick this
Pixel-perfect DOM
We capture the live DOM tree, not screen recordings. Replays are crisp at any zoom, work in your light/dark theme, and stay legible after a year of CSS changes.
AI summary on every session
A natural-language summary appears the moment a session lands — what the user did, where they got stuck, and the one error worth investigating. No more scrubbing through 15-minute replays.
Privacy by default
PII is masked in-browser before transit — 12-category regex + Luhn validation in the SDK, plus an ONNX/WebGPU GLiNER deep-scan when the Relyv extension is installed. Credit cards, emails, and SSNs never leave the browser unmasked.
Sub-3% performance overhead
Recording runs off the main thread with budget-aware mutation throttling. Real-User-Monitoring tests show under 3% Core Web Vitals impact at p95.
How it works
- 1
Drop in the SDK
A single <script> tag in <head> or one npm install. Zero config to get started; rich config for teams who want it.
- 2
Sessions stream in
Recordings start the moment the page loads, batch-uploaded with WAL-backed durability so a tab crash never costs you a session.
- 3
AI summarizes immediately
A 2-line summary, frustration score, and detected error appear within seconds of the session ending. Your inbox is sorted by impact, not chronology.
- 4
Replay with one click
Open the replay, scrub, jump to AI-flagged moments, or export a Playwright test. Browser console + network panel are right there in the player.
Under the hood
How DOM serialization beats video recording
Session replay tools that record video produce 5–50× larger files, look pixelated on retina displays, and break the moment your CSS changes. Relyv.ai serializes the DOM tree on first paint, then streams MutationObserver deltas. Replays are vector-crisp, theme-aware, and survive months of design refactors.
Console + network capture in the same timeline
Every console.log, console.error, and console.warn is captured alongside fetch/XHR requests, response status, and timing. The replay timeline shows them inline, so you see the failed POST and the resulting alert dialog in the same frame.
PII masking before data leaves the browser
A pattern-validator first-pass catches obvious PII (Luhn-valid card numbers, RFC 5322 emails, SSN shape) directly in the SDK. With the Relyv browser extension installed, an ONNX/WebGPU GLiNER model in an offscreen document does a second pass for context-bound PII (custom tokens, free-text leaks). Masked values never hit our network.
Frequently asked
How long do you store sessions?
Free: 7 days. Starter: 30 days. Pro: 90 days. Business: 1 year. Enterprise: custom (with EU/US data-residency options). Sessions are deleted on schedule, on demand, or on subject-access request.
How accurate is the replay vs. the original?
For first-party content: pixel-identical, including custom fonts, dynamic CSS, and Shadow DOM. For canvas/WebGL/video: we capture src + state, not the rendered pixels (that's a config knob). Iframes from third parties are blocked by browser security and shown as opaque blocks.
What is the performance impact?
P50 main-thread cost: under 1ms per second of recording. P95: under 3% Core Web Vitals delta on the page-under-test, measured on a Moto G Power. Recording runs off-main-thread; large mutation bursts get throttled, never dropped.
Are you GDPR + CCPA compliant?
Yes. PII is masked client-side, opt-out cookies are honoured, all subject-access requests are processable from the dashboard, and we have a signed DPA available on request. SOC 2 Type II is in progress.
How do team seats work for replay access?
All seats on a workspace share the session pool — there's no per-seat replay surcharge. Granular roles (admin / dev / viewer) gate who can export, share, or delete recordings.
Does Relyv capture mobile sessions as well as desktop?
Yes. The Wave-7 capture stack adds Pointer Events with pressure + tilt (pen / touch / mouse distinguished), a dedicated pinch-zoom gesture recorder, virtual-keyboard show/hide via the visualViewport API, plus Service Worker lifecycle, Battery API, and online/offline transitions. All 38 captured event types ship on mobile browsers — same payload, same viewer.
Can I export a Relyv replay as a single offline file?
Yes. Every session can export as one self-contained .html file with the viewer embedded inline — DOM, network, console, AI summary, and all 38 event types replay in any browser, no account, no internet. Every shared file is also a viral install for Relyv.
Can my AI agent query Relyv sessions?
Yes — Relyv is MCP-native. Claude, Cursor, and Continue call get_session_intelligence as a tool, ask "what frustrated the user in session X around timestamp Y", and get structured intent + frustration + outcome data back. No other replay tool exposes session data over MCP today.
How many event types does Relyv capture per session?
38 distinct event types covering DOM mutations, network (fetch/XHR/WebSocket/SSE/Beacon), console (log/info/warn/error/debug + uncaught + unhandledrejection), interaction (mouse + keyboard + touch + pointer + pinch + contextmenu), navigation, performance (LCP/INP/CLS/FCP/TTFB), viewport + virtual keyboard, storage (localStorage/sessionStorage/IndexedDB/cookies), visibility, form lifecycle, security signals (CSP + permission changes), drag-drop, animation, clipboard, accessibility, canvas, screen + webcam + speech, media (video/audio), environment (offline/SW/battery), and engagement (scroll-depth + active-time).
Ready to try it?
Pixel-perfect DOM reconstruction, AI summaries on every recording, and a frustration heatmap that points to the exact moment users gave up. No video files. No 1990s-style mouse trails.