PostHog vs Plausible vs Umami: Self-Hosted Privacy Analytics in 2026
Self-hosted privacy analytics compared in 2026: PostHog vs Plausible vs Umami on cost, features, performance, and the right pick for each kind of team.
You finally hit “delete” on Google Analytics. Maybe it was the GA4 migration that did it, maybe the GDPR audit, maybe just the dread of staring at another funnel UI you didn’t ask for. Now you’re standing in the empty space where your analytics used to be, and the obvious replacements all promise the same thing — “privacy-first, no cookies, you own the data.” The catch is they’re three very different tools wearing the same marketing copy.
This guide compares PostHog vs Plausible vs Umami as self-hosted privacy analytics stacks in 2026 — what each is actually optimized for, real infra cost at 1M monthly events, the trade-offs nobody puts on the landing page, and a one-page rubric for picking.
TL;DR
- Plausible — minimal, opinionated web analytics. The lightest script (under 1KB) and the cleanest UI. Picks you if you mostly want a privacy-friendly GA replacement for marketing sites.
- Umami — minimal and dead simple to self-host (single Postgres + Next.js). Picks you if you want Plausible-style simplicity for free and don’t mind a slightly less polished UI.
- PostHog — full product analytics + session replay + feature flags + experiments + LLM observability. Picks you if you’re a product team and want one tool instead of five.
The decision isn’t really “which is best” — it’s “are you doing web analytics or product analytics?” Plausible and Umami answer the first. PostHog answers the second.
What Each One Actually Is
A lot of confusion in this space comes from treating these three as interchangeable. They aren’t.
Plausible
Open-source web analytics written in Elixir, stored in ClickHouse. The hosted plan starts around $9/mo; the self-hosted Community Edition is free under AGPLv3. Plausible deliberately ships fewer features than Google Analytics — no fingerprinting, no cookies, no funnels deeper than a couple of steps — in exchange for a dashboard you can read in 30 seconds and a 1KB tracking script. Documentation lives at plausible.io/docs.
Umami
Open-source web analytics written in Next.js, stored in PostgreSQL or MySQL (ClickHouse support added in v2). MIT-licensed. The hosted version is $20/mo for unlimited websites; self-hosting is free and famously easy — one container, one database, one environment variable. Docs at umami.is/docs.
PostHog
Open-source product analytics platform written in Python/Django + TypeScript, with ClickHouse for events, Postgres for metadata, Kafka + Redis for ingestion. MIT-licensed core; some advanced features are paid in the cloud version. PostHog isn’t trying to be GA — it’s trying to be Mixpanel + Hotjar + LaunchDarkly + Amplitude in a single self-hostable bundle. Docs at posthog.com/docs.
Head-to-Head Comparison
| Dimension | Plausible | Umami | PostHog |
|---|---|---|---|
| Primary focus | Web analytics | Web analytics | Product analytics + flags + replay |
| License | AGPLv3 (CE) | MIT | MIT (core) |
| Database | ClickHouse | Postgres / MySQL / ClickHouse | ClickHouse + Postgres + Kafka |
| Tracking script size | ~1 KB | ~2 KB | ~50 KB (full SDK) / 9 KB (lite) |
| Cookies needed? | No | No | No (configurable) |
| Session replay | No | No | Yes |
| Feature flags | No | No | Yes |
| Funnels | Limited | Basic | Full |
| Heatmaps | No | No | Yes |
| LLM observability | No | No | Yes (PostHog AI) |
| GDPR-compliant out of box | Yes | Yes | Configurable |
| Self-hosting difficulty | Medium | Easy | Hard (multi-service) |
| Min self-host RAM | ~1 GB | ~512 MB | ~4 GB (recommended 8+) |
A few of those rows deserve unpacking, since they’re where teams pick wrong.
Tracking Script Weight
Plausible and Umami win on script size by a wide margin. If your conversion rate is sensitive to Largest Contentful Paint — marketing sites, e-commerce landing pages — that 1-2KB script versus PostHog’s 50KB main bundle matters. PostHog ships a lite “snippet” loader and code-splits autocapture/session-replay/flags, but the full SDK is genuinely heavy because it’s doing genuinely more.
Self-Hosting Difficulty
Umami’s docker-compose is famously a five-minute install. Plausible’s CE is a docker-compose that includes ClickHouse — slightly more moving parts but still well-documented. PostHog’s official self-host path is a Kubernetes helm chart with ClickHouse, Postgres, Kafka, Redis, and a worker fleet. It runs fine, but you are now operating a small data platform.
Real Infra Cost at 1M Events / Month
This is the part nobody publishes honestly. Approximate monthly cost on Hetzner Cloud to run each self-hosted at 1M pageviews or events/month, including the database, with daily snapshots:
| Stack | Recommended VM | Approx. EUR/month |
|---|---|---|
| Umami (single container + managed Postgres) | CPX21 (3 vCPU, 4 GB) + 1 GB Postgres | ~€11 |
| Plausible CE (app + ClickHouse) | CPX31 (4 vCPU, 8 GB) | ~€16 |
| PostHog self-host (k3s) | 2× CPX41 (8 vCPU, 16 GB) | ~€60 |
Compare those to PostHog Cloud’s free tier (1M events/month free) or Plausible Cloud’s $9-19/mo plans. For most teams under 5M events/month, hosted is cheaper than self-host once you price in your own ops time. Self-host wins on cost only at scale, or when “data must stay on our infra” is a hard requirement.
Privacy & Compliance
All three avoid the biggest GA-era footgun: dropping a third-party cookie that requires a consent banner. None of them set cookies by default. None send IP addresses to a foreign server when self-hosted.
- Plausible: uses a daily-rotating hash of (IP + user agent + domain + salt) to derive a “visitor ID,” then discards both inputs. EU-friendly out of the box.
- Umami: similar hash-based visitor ID. Server-side hashing only.
- PostHog: configurable. The default
posthog-jsSDK uses localStorage to persist a distinct ID, which in some EU jurisdictions still requires consent. PostHog has a GDPR-mode flag that switches to memory-only IDs.
If you sell into Germany or France specifically, do a real DPIA before committing to PostHog without configuration — but all three are dramatically friendlier than GA4 under EU case law (notably the Austrian and French DPA rulings against Google Analytics).
Pros and Cons
Plausible
Pros
- Cleanest dashboard in the category — non-technical stakeholders read it without training
- 1KB script — measurably better Core Web Vitals impact than GA4 or PostHog
- Built-in goal conversions, UTM tracking, custom events without complexity tax
- Solid CSV/API export
Cons
- AGPLv3 — fine for SaaS, awkward for some enterprise legal teams
- No session replay, no heatmaps, limited funnels
- Self-host upgrades occasionally require ClickHouse schema migrations
Umami
Pros
- Easiest self-host in the category — single container + Postgres
- MIT license is friendlier for embedding
- Multi-site support without per-site pricing
- Genuinely free in self-host with no feature gates
Cons
- UI is more spartan than Plausible — less polished for non-technical viewers
- Smaller plugin / integration ecosystem
- Funnel and retention analysis weaker than PostHog
PostHog
Pros
- One platform for analytics + session replay + feature flags + experiments + surveys + LLM observability
- Best-in-class funnel and retention analytics
- SQL access via the dashboard for ad-hoc queries
- Generous cloud free tier (1M events/month)
Cons
- Heavy self-host — multi-service, real ops burden
- Tracking script much larger; harder to keep marketing-site CWV green
- More configuration required to be GDPR-default
Who Should Use Which
Use Plausible if: you run a marketing site or content site, you want one number per page per day, you care about page weight, and your stakeholders shouldn’t need a tutorial. Pairs naturally with the static-site stacks I compared in Astro vs Next.js for content sites.
Use Umami if: Plausible looks right but you want MIT licensing or the absolute minimum self-host complexity. Also a good fit when you operate dozens of small sites — Umami’s multi-site model has no per-site pricing.
Use PostHog if: you build a product, not a site. If you’re shipping a SaaS app and you’d otherwise buy Mixpanel + LaunchDarkly + Hotjar, PostHog replaces all three. The heavier script and infra cost pay for themselves the moment you actually use feature flags and session replay. If you’re already using PostHog for LLM observability, see how it fits with patterns from Claude API prompt caching.
A One-Page Decision Rubric
Answer these in order. The first “yes” picks the tool.
- Do you need session replay, feature flags, or A/B experiments? → PostHog
- Is the property a SaaS app with logged-in user journeys you need to analyze? → PostHog
- Is it a marketing/content/docs site and you want maximum polish? → Plausible
- Same as 3, but you want MIT license or the dead-simplest self-host? → Umami
- Otherwise → start with Plausible Cloud at $9/mo; revisit when you outgrow it.
FAQ
Can I run all three side-by-side?
Technically yes — they don’t conflict. In practice, two tracking scripts means duplicated script weight and confused stakeholders. Pick one for source of truth.
Which one handles SPAs (React/Vue/Svelte) best?
All three support SPA route-change tracking, but PostHog autocaptures by default. Plausible and Umami need an explicit trackPageview() call on route change — covered in their docs.
Is self-hosted PostHog actually production-viable for a small team?
Yes, but budget real time. The official Helm chart works, but you’ll spend a weekend per quarter on ClickHouse housekeeping, Kafka monitoring, and upgrade runs. Many small teams use PostHog Cloud with EU region instead to satisfy data-residency without the ops tax.
Do any of them work without JavaScript?
Plausible and PostHog both offer server-side / proxy event ingestion APIs you can call from your backend. Umami also supports a /api/send endpoint. Useful for tracking conversions where the client may not load JS (email opens, server-rendered confirmations).
What about Matomo / Fathom?
Both are legitimate options too. Matomo is the most feature-complete GA clone but ships with a heavier UI; Fathom is hosted-only (no self-host) but exceptionally simple. They didn’t make the cut here only because the three above cover the typical 2026 decision tree.
Bottom Line
PostHog vs Plausible vs Umami isn’t really one debate — it’s two. If you’re picking a web analytics tool, the choice is between Plausible (polish + price) and Umami (simplicity + MIT). If you’re picking a product analytics tool, PostHog is the only realistic open-source contender against the Mixpanel/Amplitude tier. Self-host all three only if compliance demands it; otherwise the hosted tiers are cheap enough that your operations time is the more expensive line item.
If you’re building the rest of the stack around this decision, the Cloudflare Workers vs AWS Lambda comparison covers where to host the proxy layer, and Drizzle vs Prisma covers the database side if you go the Umami-on-Postgres route.
Product recommendations are based on independent research and testing. We may earn a commission through affiliate links at no extra cost to you.