Langfuse logo

Langfuse Review (2026)

The open-source LLM observability default. Free, MIT-licensed, genuinely feature-complete when self-hosted - now a ClickHouse subsidiary after its January 2026 acquisition.

Hands-on tested

Rating

5.0

Starting Price

$29/mo

Free Plan

Yes

SDKs & Frameworks

7

Deployment

4

Best For

Teams that want an open-source observability platform they can self-host without losing features

Last Updated:

10 Things You Should Know About Langfuse

  1. 1 Acquired by ClickHouse on 16 January 2026; MIT license and self-host preserved
  2. 2 Only 3 features are enterprise-gated in self-host - tracing, evals, prompt management and RBAC are all free
  3. 3 Roughly $101/mo at 1M events versus LangSmith's ~$2,514/mo for comparable volume
  4. 4 Self-host v3 requires Postgres, ClickHouse, Redis and S3-compatible storage
  5. 5 Founded 2022 in Berlin, YC W23; had 2,000+ paying customers at acquisition

Pros & Cons

Pros

  • Self-hosting is free and genuinely feature-complete - only 3 features are enterprise-gated
  • Roughly 25x cheaper than LangSmith at high trace volume
  • Framework-agnostic - works with any stack, not locked to one
  • Maps the OpenTelemetry GenAI semantic conventions
  • Generous discounts for startups, students, nonprofits and OSS projects

Cons

  • Self-host now needs Postgres plus ClickHouse, Redis and S3 - four moving parts
  • The v3 ClickHouse migration is where self-hosters get stuck
  • Now owned by ClickHouse, which raises a long-term strategic-ownership question
  • Full regression testing needs you to assemble the orchestration yourself
  • gRPC OTLP not supported yet, HTTP only

Features

Tracing and observability for LLM apps
Evals with LLM-as-judge and custom scorers
Prompt management and versioning
Human annotation queues
OpenTelemetry backend on /api/public/otel
Full self-hosting under MIT

What Langfuse actually is

Langfuse is the open-source default for LLM observability. It does four things - tracing, evals, prompt management, and human annotation - and it does them without locking you to any particular framework. If you’ve asked “what’s the self-hostable alternative to LangSmith,” the answer you got was almost certainly Langfuse.

As of January 2026 it’s owned by ClickHouse, acquired alongside ClickHouse’s $400M Series D at a $15B valuation. At the time it had over 2,000 paying customers and was used by 19 of the Fortune 50. ClickHouse committed to keeping the MIT license, self-hosting, and the roadmap intact - so far that’s held, but it’s a fact worth filing away when you’re choosing a platform for the next three years.

The reason people pick it: self-host that doesn’t cripple you

This is the whole story, so it goes first. Most “open-source” observability tools give you a hobbled core and gate everything useful behind a paid cloud tier. Langfuse doesn’t.

In the self-hosted build, only three features are enterprise-gated: Organization Creators, the Instance Management API, and UI customization. Everything you actually need - tracing, evals, prompt management, human annotation, RBAC - is free under MIT. Compare that to LangSmith and Braintrust, where self-hosting is an Enterprise-only, contact-sales affair.

That’s the differentiator. You can run the real product, for free, on your own infrastructure, with your data never leaving your cloud.

Pricing: cheap, and dramatically cheaper at scale

TierPriceIncludedOverageRetention
Hobby$050k units/mo (hard cap)none30 days
Core$29/mo100k units$8/100k90 days
Pro$199/mo100k units$8/100k3 years
Enterprise$2,499/mo100k units$8/100kcustom

The quota unit is “billable units” - traces plus observations plus scores - and the overage rate steps down from $8 to $6 per 100k at higher volume.

But the number that matters is the one at scale. At 1M events per month, Langfuse runs about $101/mo. LangSmith, for comparable volume, runs about $2,514/mo - roughly 25x more. Self-host and the per-trace cost disappears entirely. This is the single most-cited reason teams migrate off LangSmith, and it’s not close.

There are also real discounts: 50% off year one for startups, up to 100% off for students and researchers, and credits for nonprofits and OSS projects.

Self-hosting: what you actually get, and the catch

You get the full product. The catch is operational, and it’s a genuine one.

Langfuse v3 moved from “just Postgres” to Postgres plus ClickHouse plus Redis plus S3-compatible storage. That’s four services to stand up and keep running. The migration to that architecture is where self-hosters get stuck - GitHub has a visible string of “ClickHouse migration mess” threads (#7490, #9604, #12668) reporting memory errors, timeouts and duplicate-column errors, with clusters needing auto-migration disabled and run manually.

So the honest framing: the software is free, the ops time is not. If you have a team that can run a four-service stack, this is the best deal in the category. If you don’t, the free cloud Hobby tier or the $29 Core plan sidesteps the whole problem.

Langfuse versus LangSmith

The clean comparison. LangSmith is turnkey but expensive and coupled to the LangChain stack; self-hosting is Enterprise-only. Langfuse is framework-agnostic, radically cheaper, and self-hostable for free - but you assemble more of the orchestration yourself, especially for full regression testing where Braintrust is more turnkey.

If you’re already all-in on LangChain and LangGraph and you don’t mind the bill, LangSmith’s integration is tighter. For everyone else, Langfuse is the better default.

Should you use it?

Use Langfuse if you want an open-source observability platform you can self-host without losing features, or you want the cheapest managed option at scale, or you simply don’t want to be locked to one framework’s ecosystem.

Think twice if you can’t spare the ops capacity for a four-service self-host and you’re at high enough volume that the managed tiers get expensive - though even then, it’s cheaper than the alternatives.

Bottom line: it’s the open-source default because it earns the title. Just go in clear-eyed that “free self-host” means free software, not free operations.


Pricing and features verified against langfuse.com on 23 July 2026. This category ships breaking changes monthly - we re-verify every 30 days.

Pricing Plans

Hobby

$0

  • 50k billable units per month, hard limit
  • 30-day retention, 2 users
  • Community support
  • Or self-host free under MIT
Most Popular

Core

$29/mo

  • 100k units, then $8 per 100k
  • 90-day retention, unlimited users
  • In-app support

Pro

$199/mo

  • 3-year retention
  • SOC2 and ISO27001 reports
  • BAA available

Enterprise

$2,499/mo

  • Audit logs, SCIM, SLA
  • Custom rate limits
  • Dedicated engineer

SDKs & Frameworks

Python SDK JS / TypeScript SDK LangChain LlamaIndex OpenAI SDK Vercel AI SDK OpenTelemetry (OTLP)

Deployment

Cloud (free Hobby tier) Self-hosted, MIT licensed Docker Compose Kubernetes

Eval Methods

LLM-as-a-judge Custom scorers Dataset management Human annotation queues

Our Verdict

The open-source default for good reason - self-hosting is free and doesn't cripple the product, and it's dramatically cheaper than LangSmith at scale. The catch is operational, not commercial - the v3 architecture is four services now, and the migration bites. If you have the ops capacity, nothing else in the category gives you this much for free.

Similar Tools

Frequently Asked Questions

Is Langfuse really free to self-host?

Yes, and unusually so. The self-hosted build is MIT-licensed and free, and only three features are enterprise-gated - Organization Creators, the Instance Management API, and UI customization. Core tracing, evals, prompt management, human annotation and RBAC are all included free. That's far more generous than LangSmith or Braintrust, where self-host is Enterprise-only.

What changed when ClickHouse acquired Langfuse?

Officially, nothing about the product - ClickHouse committed to keeping the MIT license intact, self-hosting as a first-class option, and the roadmap unchanged. The real change is strategic - Langfuse is now a subsidiary of a database company, which is worth weighing if you're betting a multi-year stack on it.

Does Langfuse support OpenTelemetry?

Yes. Langfuse operates as an OpenTelemetry backend, receiving traces on an OTLP endpoint, and aims to comply with the GenAI semantic conventions. It supports OTLP over HTTP/JSON and HTTP/protobuf. gRPC is not supported yet.

Why is Langfuse so much cheaper than LangSmith?

At high volume the gap is enormous - around $101/mo at 1M events versus LangSmith's roughly $2,514/mo for comparable traces. Langfuse bills "billable units" at $8 per 100k with graduated discounts, and self-hosting removes the per-trace cost entirely. It's the single most-cited reason teams switch.

What's the catch with self-hosting Langfuse?

Operations. The v3 architecture moved from "just Postgres" to Postgres plus ClickHouse, Redis and S3-compatible storage. That's four services to run, and the migration is where people get stuck - GitHub has a string of migration threads reporting memory errors and timeouts. The software is free; the ops time is not.

Related Articles

guide

The Langfuse Free Tier Explained in 2026 - Limits, Cap and When to Leave

Langfuse has two free options, and they are not the same. Here's what the Hobby cloud tier's 50k-unit cap really counts, how the free MIT self-host differs, and when Opik's free tier beats both.

July 26, 2026

how-to

How to Integrate Langfuse with LangChain in 2026 - A Practical Guide

Wire Langfuse tracing into a LangChain or LangGraph app with a callback handler, see every chain and tool call in the dashboard, and add evals - plus the self-host gotcha and when Opik is the cheaper managed pick.

July 26, 2026

guide

Langfuse Pricing Explained (2026) - What You Actually Pay

Langfuse bills "billable units" - traces plus observations plus scores - at $8 per 100k, and self-host is free under MIT. Here is how the meter really works, a worked bill estimate, and when a cheaper pick beats it.

July 26, 2026

comparison

Langfuse vs Arize Phoenix in 2026 - License vs Eval Depth

The two open-source LLM observability defaults, compared honestly. Langfuse wins on license clarity and cheap self-host, Phoenix wins on RAG eval and OpenTelemetry-native architecture. Here is which one fits which team.

July 26, 2026

comparison

Langfuse vs Braintrust 2026 - Open Self-Host vs Turnkey Evals

Langfuse is the cheap, open, self-hostable observability default. Braintrust is the most turnkey eval and regression-testing platform, with an uncapped billing meter. Here is the honest split by use case, plus where Opik fits.

July 26, 2026

comparison

Langfuse vs Datadog for LLM Observability (2026) - An Honest Head-to-Head

Datadog LLM Observability puts your traces in the same pane of glass as your infra, logs and APM. Langfuse is open-source, self-hostable and LLM-specialized. This is a neutral comparison - the pricing model, the self-host reality, the eval depth - with a clear pick for each kind of team.

July 26, 2026

comparison

Langfuse vs Helicone in 2026 - Why One of These Is a Dead End

Both are open-source LLM observability tools you can self-host for free. But Helicone went into maintenance mode after Mintlify bought it in March 2026, and that decides most of this comparison. Here is the honest split, by use case.

July 26, 2026

comparison

Langfuse vs Helicone vs Opik in 2026 - And Why One Is Off the Table

Three open-source observability tools compared. Langfuse is the MIT default, Opik is the cheapest cloud with the most permissive license, and Helicone is in maintenance mode after its acquisition. Here is the honest pick.

July 26, 2026

comparison

Langfuse vs LangSmith in 2026 - The Honest Head-to-Head

LangSmith is the most turnkey observability for LangChain apps and roughly 25x more expensive than Langfuse at 1M traces. Langfuse is open-source, self-hostable free, and framework-agnostic. Here is the real split, by use case.

July 26, 2026