Best open-source session replay tools (2026)

Every fact below was read from each project's own repo, docs or pricing page on 28 July 2026 — licenses from LICENSE files, limits from the vendors' own documentation. Nothing comes from a rival's comparison post.

The short answer

Library, product, or suite — the distinction that sorts the category

"Open-source session replay" hides three different purchases:

  1. A library (rrweb) gives you capture and playback as code. Storage, ingestion, search, identity, privacy governance and hosting the player are yours to build.
  2. A product you host (OpenReplay) gives you the finished tool — recordings list, DevTools, funnels — and hands you the server: provisioning, upgrades, backups, scaling, retention.
  3. A suite you mostly don't host (PostHog) is open source at the core, but the maintained experience is their cloud, priced per use with a generous free floor.

Decide which of the three you're actually shopping for and most of the arguing disappears.

At a glance

Open-source session replay projects compared
Project What's genuinely free What you still run or build the pivot Replay data kept
OpenReplay All core features, self-hosted; cloud free tier: 1,000 sessions/mo, 2 users An Ubuntu 24.04 box — their stated minimum is 2 vCPUs, 8 GB RAM, 50 GB — plus upgrades, backups, scaling Self-host: your disk · cloud Free: 7 days · Serverless: 30 days
PostHog 5,000 web recordings/mo on cloud, every month, no card Nothing — it's SaaS in practice; self-host is discouraged by the vendor 30 days free · 90 days pay-as-you-go · up to 1 year on $250–750/mo platform packages
rrweb The entire library — record, replay, player UI (~20k GitHub stars, July 2026) Storage, ingestion, search, identity, privacy governance, player hosting — everything after the emit callback Wherever you store events — entirely yours
highlight.io Standalone signup still exists Its own homepage banner now says "Migrate your Highlight account to LaunchDarkly" Per LaunchDarkly Observability going forward

1. OpenReplay — the self-host anchor

Free self-hosted (AGPL v3 core) · cloud: free to 1,000 sessions/mo, then Serverless at $5.95 per 1,000 sessions with 30-day retention, or Dedicated instances from $199/mo · 12.3k GitHub stars · v1.27.0 shipped May 2026, repo active.

The genuine article: session replay, DevTools context (network, console, JS errors, state), funnels and heatmaps, co-browsing over WebRTC (Assist), a Chrome-extension bug recorder (Spot), an MCP server, and Android/iOS/React-Native SDKs — with the core self-hostable and "everything OpenReplay captures stays in your cloud" (their README). Privacy defaults are strong: input values are obscured by default in their sanitization docs.

The license, precisely, because most pages get it wrong: the monorepo defaults to AGPL v3, the ee/ directory (SSO/SAML, SCIM, roles, audit trail) is under a proprietary enterprise license, and the tracker npm package is MIT.

What "free" costs here. Their own deployment docs state the minimum: Ubuntu Server 24.04 LTS, "2 vCPUs, 8 GB of RAM, 50 GB of storage" — and that's "enough for a low/moderate volume." You patch it, back it up, monitor it and scale it. On their cloud, watch the pricing page's tabs: it opens on Dedicated at $199/mo, and the free tier (1,000 sessions/mo, 7-day retention) and the $5.95-per-1,000-sessions Serverless plan sit behind the Serverless tab.

The capability line that matters for a store: no experimentation product. Feature flags appear only in OpenReplay's archived v1.17.0 docs; the current docs have no feature-flag or experiments section at all. OpenReplay's own positioning is "session replay for developers" — reproducing issues, not moving conversion.

Choose OpenReplay if data ownership is the requirement — compliance, contracts, or principle — and you have an engineer who'll own the box.

2. PostHog — open-source-rooted, best used hosted

Cloud: 5,000 web recordings free every month, then from $0.005 per recording · replay retention 30 days on free, 90 days pay-as-you-go, up to 1 year on the $250–750/mo platform packages · core MIT.

PostHog belongs on this list with an asterisk it supplies itself: the core is MIT and a Docker Compose deployment exists, but its own docs call self-hosted deployments "officially unsupported" and say "PostHog Cloud is far and away the best experience for the vast majority of our users." Treat it as the hosted option with open-source roots, not the self-host option.

What you get is the widest surface here: replay (web, plus GA mobile replay on four platforms), product analytics, heatmaps via its toolbar, error tracking, surveys — and, unlike everything else on this page, a real experimentation product, Bayesian and frequentist, billed through feature-flag requests with 1M requests free monthly. The honest caveats are its own docs': the no-code web experiments editor is beta, and a billing limit you set is destructive — "if you exceed the billing limit you set, your additional data is lost forever."

Choose PostHog if you want free-at-small-scale replay plus experiments and your team is comfortable in a developer-first tool.

3. rrweb — the engine, not the car

MIT · about 20,000 GitHub stars (July 2026) · 2.1.1 shipped 21 July 2026 · maintained by a three-person core team; the maintainers also sell a hosted backend, rrweb Cloud.

rrweb is what a large share of commercial replay is built on — its own README's adopter wall lists PostHog, Sentry, Datadog, Mixpanel, Amplitude, New Relic and Pendo. Four packages do the work: snapshot (serialize the DOM), record ("builds on an initial snapshot to record all HTML state changes (mutations) and user interactions"), replay, and a player UI.

The contract is one line in their guide: the recorder emits events and "all you need to do is store the emitted events in any way you like." That sentence is the entire build-vs-buy decision. The library ships no storage, no ingestion pipeline, no session list, no search, no identity, no retention controls, no governance UI. Masking primitives exist — rr-block/rr-ignore/rr-mask classes and custom mask functions — but the defaults are minimal: maskAllInputs is false; only password inputs are masked out of the box.

Choose rrweb if you're building replay into a product of your own. For running a store, it isn't a tool — it's a project.

4. A note on highlight.io

It was the fourth standalone option. In April 2025 the team announced "We're joining LaunchDarkly!", and its homepage now banners "Migrate your Highlight account to LaunchDarkly." The repo is still active and standalone signup still works, so it isn't dead — but starting a new deployment on a product with an official migration path off itself is a bet this page can't recommend.

The build-vs-buy maths, honestly

No invented dollar figures — the inputs are all first-party, and your own payroll does the rest:

Building on rrweb means building: an ingestion endpoint and batching strategy, storage and its growth curve, a session index and search, identity stitching, retention and deletion mechanics, privacy governance beyond password-masking defaults, and hosting for the player. The maintainers' own roadmap points "storage engine" at their commercial cloud — the OSS library stores nothing by design. Then it's yours to maintain through every major version.

Self-hosting OpenReplay means running: an 8 GB / 2 vCPU / 50 GB box as the stated floor ("low/moderate volume"), plus upgrades, backups, monitoring and scale-out as traffic grows. The software is free; the pager is not.

Hosted free tiers mean limits you should read as product decisions, not tricks: 1,000 sessions and 7-day retention on OpenReplay's free cloud; 5,000 recordings and 30-day retention on PostHog's. Generous for a side project; on a store doing real traffic during a peak week, they're a sampling policy you didn't choose.

The right question isn't "free or paid." It's whose engineers run the pipeline, and what happens to the finding once you have it — because none of the three open-source options can run the A/B test its own replay suggests (PostHog can, in its own developer-first way; rrweb and OpenReplay cannot at all).

This is the vendor section. It's labelled, it's last, and the table above didn't rank it.

Mtrix is not open source. Its recorder and player are built fully in-house — a complete snapshot of the DOM, every mutation after it, and the site's CSS extracted and stored — so a replay reproduces what the visitor actually saw. We built the storage, the pipeline, the search and the governance behind it, which is exactly the "other 90%" this page keeps pricing — so we'll say it plainly: it's real work, and for many teams it's the wrong work to own.

What buying it gets a DTC brand that a self-hosted pipeline doesn't:

  • Replay tied to money and to change. Sessions are searchable by page, event, error and purchase; filterable by experiment and variant; and the experiment engine that acts on what you saw is in the same product — four test types plus bandit allocation, no cap, first test live in about five minutes.
  • 100% capture with rage-click, dead-click and form-abandonment detection.
  • Governed privacy, not class names. Password, card, CVV, SSN and auth-token fields are detected by type and name and masked before they leave the browser; block/mask/unmask attributes inherit to child elements.
  • A processor relationship and region choice — hosting region is customer-selectable, EU by default. SOC 2 Type II and ISO 27001 certified.
  • Retention as policy: replays kept 1 month; events and purchases 3 years.

Who shouldn't buy it: if free or self-hosted is the requirement, use OpenReplay — genuinely. If 5,000 recordings a month covers you, PostHog's free tier is honest and excellent. Mtrix is platform-tier, has no free plan (one month free, no credit card, then paid), and earns its price only where a percentage point of conversion is worth more than the subscription.

FAQ

Is there a truly free open-source session replay tool?
Yes — two, in different senses. OpenReplay's AGPL core is free to self-host with all core features; the cost is the server and the engineer who owns it. PostHog's cloud gives 5,000 recordings a month free, every month, no card — with 30-day retention. rrweb is free as a library, but it's a component, not a tool.
What is rrweb?
The MIT-licensed library that records DOM snapshots and mutations and replays them — the engine under much of the commercial replay category (its README's adopter wall lists PostHog, Sentry, Datadog, Mixpanel, Amplitude, New Relic and Pendo). It ships no storage, search or dashboard: you build the product around it.
What does it cost to self-host session replay?
OpenReplay's own docs state the minimum: an Ubuntu 24.04 machine with 2 vCPUs, 8 GB RAM and 50 GB storage — "enough for a low/moderate volume" — scaled up from there as traffic grows. The real line item is the engineer-hours: upgrades, backups, retention, scaling, and privacy configuration are all yours.
Is session replay GDPR-compliant if I self-host it?
Self-hosting changes where the data lives, not whether you may record it — consent and masking obligations apply either way. Defaults differ sharply between tools: rrweb masks only password inputs out of the box, while OpenReplay obscures input values by default. The full picture — what must be masked, when consent is needed, and how Mtrix handles it — is in our guide: session replay, privacy and GDPR.
What's the difference between a session replay library and a session replay tool?
A library (rrweb) is code that captures and replays; everything operational — storage, search, identity, retention, governance — is your build. A tool (OpenReplay, PostHog, Mtrix) is the finished loop: record, find, watch, act. If your team ships stores rather than infrastructure, you want a tool.

If you costed it out and landed on "buy"

Tomas Reinart Ecommerce performance Writes about store performance and the DTC tool stack at Mtrix.
Build with Mtrix