Key Takeaways
- Langfuse (langfuse.com) is an open-source LLM observability and evaluation platform, built on the ClickHouse analytical database
- Core codebase is MIT licensed; a smaller set of Enterprise Edition (EE) features in
ee/-prefixed directories carry a separate commercial license defined in the repo's ownee/LICENSEβ this is why GitHub labels the overall license "Other," not plain MIT - Self-hosting (Docker Compose, Kubernetes/Helm, or AWS/Azure/GCP infrastructure-as-code templates) is free regardless of any Cloud tier
- Langfuse Cloud (the managed option) has a free Hobby tier and paid tiers at $29, $199, and $2,499 per month
- Over 34,700 GitHub stars and roughly 3,794β3,800 forks; not archived, and shipping new releases at a near-daily pace
- As of January 2026, Langfuse is part of ClickHouse, per the repository's own LICENSE file copyright header ("Copyright (c) 2023-2026 ClickHouse, Inc.")
- Founded in 2023 and went through Y Combinator's Winter 2023 (W23) batch
π In One Sentence
Langfuse is an open-source LLM engineering platform for tracing, evaluating, and debugging AI applications, with an MIT-licensed core codebase, a free self-hosting path, and a separate managed Langfuse Cloud with tiers from free to $2,499/month; as of January 2026 it is part of ClickHouse.
π¬ In Plain Terms
Langfuse watches what your AI application does β every call to a model, every chain of steps, every user session β so you can see what happened, score whether the output was good, and test prompt changes against real examples. You can run it yourself for free, or let Langfuse host it for you at a range of prices.
πNote: This review is the deep-dive companion to Langfuse's entry in the Local LLM Software Directory β see that page for how Langfuse compares at a glance to other local AI tools.
What Is Langfuse?
Langfuse is an open-source platform for tracing, evaluating, and debugging LLM applications in production, built on the ClickHouse analytical database under the hood. Its own GitHub description reads: "Open source agent evals & observability: Trace, evaluate, and improve LLM applications with one open platform."
- Product type: an LLM engineering / observability platform, not an inference engine and not a model β it observes and evaluates calls your application makes to other LLM providers
- Underlying database: ClickHouse, the analytical database that Langfuse itself is now organizationally part of
- Repository created: 2023-05-18; the project went through Y Combinator's Winter 2023 (W23) batch
- Ownership: as of January 2026, Langfuse is part of ClickHouse, Inc. β confirmed by the copyright header in the GitHub repository's own LICENSE file ("Copyright (c) 2023-2026 ClickHouse, Inc.")
- Scale: over 34,700 GitHub stars and roughly 3,794β3,800 forks; not archived, with releases published almost daily
Project History
Langfuse's GitHub repository was created on 2023-05-18, and the project went through Y Combinator's Winter 2023 (W23) accelerator batch. As of January 2026, Langfuse is part of ClickHouse, Inc., per the copyright header in the repository's own LICENSE file. The project ships new releases at a fast, near-daily pace β for example, v4.38.0 was published 2026-09-17 and v4.37.0 the day before, on 2026-09-16 β indicating genuinely active, continuously maintained development rather than a stalled or abandoned project. A detailed, dated version-by-version changelog was not independently verified for this review, so no numbered milestone list is presented here; check Langfuse's own GitHub releases page for the full history.
What Does Langfuse Actually Do?
Langfuse's feature set centers on watching, scoring, and iterating on LLM application behavior in production, per its own site and documentation.
- LLM application observability / tracing β trace and inspect complex logs, chains, and user sessions so you can see exactly what your application did on any given request
- Prompt management β centralized version control for prompts, with caching, so you can iterate on prompt text without redeploying application code
- Evaluations β LLM-as-a-judge scoring, code-based evaluators, and collection of end-user feedback, so output quality can be measured rather than guessed at
- Datasets β build test sets and benchmarks for continuous evaluation as you change prompts or models
- LLM Playground β an in-UI space for interactively testing and iterating on prompts without a separate script
- API and SDKs β an OpenAPI-specified API plus official Python and JavaScript/TypeScript SDKs
- Framework integrations β works with the OpenAI API, LangChain, LlamaIndex, and other frameworks
Usage Examples: Three Ways to Use Langfuse
These are concrete workflows built from Langfuse's documented feature set above.
Install and Run Langfuse
Langfuse is a self-hosted web platform, not a downloadable end-user binary. Per its own documentation, self-hosting options include Docker Compose for local or quick setup (roughly a 5-minute setup per the docs), a single-VM deployment, Kubernetes via a Helm chart, and infrastructure-as-code templates for AWS, Azure, and GCP. Langfuse Cloud is a separate, optional managed alternative to self-hosting.
Source | Link |
|---|---|
| Official site | langfuse.com |
| Documentation (self-hosting, SDKs, API) | langfuse.com/docs |
| GitHub repository (source code, Docker Compose files) | github.com/langfuse/langfuse |
| Langfuse Cloud signup (managed, free Hobby tier) | langfuse.com |
Verify the current recommended self-hosting method (Docker Compose vs. Kubernetes/Helm vs. a cloud IaC template) directly on langfuse.com/docs before deploying, since setup instructions can change between releases.
Langfuse Pricing
Self-hosting Langfuse is free and open source regardless of the Cloud tiers below β those tiers apply only to the managed Langfuse Cloud product. Self-host with Docker Compose locally or Kubernetes/Helm/cloud infrastructure-as-code templates in production, at no cost, using the MIT-licensed core.
Self-hosted (any deployment method)
- Price:
- Free, open source (MIT core)
- Key limits:
- No Langfuse-imposed usage cap; your own infrastructure is the only constraint
Cloud Hobby
- Price:
- $0/month
- Key limits:
- 50k units/month included, 30 days data retention, 2 users, community support
Cloud Core
- Price:
- $29/month
- Key limits:
- 100k units/month included ($8 per additional 100k), 90 days retention, unlimited users, in-app support
Cloud Pro
- Price:
- $199/month
- Key limits:
- 100k units/month included ($8 per additional 100k), 3 years retention, unlimited annotation queues, high rate limits, SOC2/ISO 27001 reports, HIPAA available
Cloud Enterprise
- Price:
- $2,499/month
- Key limits:
- Everything in Pro plus Teams features, custom rate limits, uptime and support SLA, named lead support engineer
Figures above are Langfuse's own published Cloud pricing as of this review β verify current tiers and limits directly on langfuse.com/pricing before budgeting, since pricing pages can change.
Langfuse vs. Helicone
Helicone (helicone.ai) is another well-known open-source LLM observability tool, and the two are often compared. PromptQuorum has not yet published a dedicated Helicone review, so this comparison is based on Helicone's own site rather than a hands-on PromptQuorum test β verify current details directly on helicone.ai before deciding.
Self-hosting model
- Langfuse:
- Docker Compose, Kubernetes/Helm, or cloud IaC templates
- Helicone:
- Also offers a self-hosted deployment path β verify current setup steps on helicone.ai
License / openness
- Langfuse:
- MIT core, with a separate commercial license for a smaller
ee/-prefixed feature set - Helicone:
- Also open source β verify its exact current license terms on helicone.ai rather than assuming they match Langfuse's
Pricing model
- Langfuse:
- Free self-hosted; managed Cloud from $0 (Hobby) to $2,499/month (Enterprise)
- Helicone:
- Has its own separate pricing structure β check helicone.ai/pricing directly
Evaluation features
- Langfuse:
- LLM-as-a-judge, code-based evaluators, datasets, user feedback collection
- Helicone:
- Also offers evaluation-related tooling β verify current feature scope on helicone.ai
Primary focus
- Langfuse:
- Full LLM engineering platform: tracing, evals, prompt management, datasets, playground
- Helicone:
- LLM observability and gateway-style request logging, per its own positioning
This table intentionally hedges on Helicone-specific details PromptQuorum has not independently verified β treat the Langfuse column as this review's verified facts and the Helicone column as a starting point for your own comparison on helicone.ai.
Who Should Use Langfuse?
Whether Langfuse fits depends on whether you are running LLM applications in production and need visibility into what they are actually doing, rather than just building or prototyping one.
Competitors and Alternatives
No dedicated LLM-observability-tool review exists yet on PromptQuorum, so the links below are adjacent LLM production-tooling reviews rather than direct observability competitors β useful context for readers evaluating Langfuse alongside the rest of their LLM stack. For direct observability competitors, Helicone (see the comparison above), PromptLayer, and Arize Phoenix are worth researching by name on their own sites.
- LiteLLM Review β an LLM gateway/router that sits between your application and multiple model providers; often used alongside an observability tool like Langfuse rather than instead of it.
- LangChain Review β the framework Langfuse most commonly instruments via its callback handler and SDKs.
- Dify Review β a broader LLMOps/workflow-builder platform with its own tracing and evaluation features, for readers comparing an all-in-one platform against a dedicated observability layer.
Common Mistakes When Evaluating Langfuse
Most confusion about Langfuse comes from its mixed license, its relationship to ClickHouse, or conflating it with an inference engine.
Frequently Asked Questions
What is Langfuse?
Langfuse (langfuse.com) is an open-source LLM engineering platform for tracing, evaluating, and debugging AI applications in production, built on the ClickHouse analytical database. Its own GitHub description calls it "open source agent evals & observability."
Is Langfuse free?
Self-hosting Langfuse is free and open source (MIT core), regardless of Cloud pricing tiers. Langfuse Cloud, the managed hosted option, also has a free Hobby tier, plus paid tiers at $29, $199, and $2,499 per month.
Is Langfuse open source? What license does it use exactly?
Langfuse's core codebase is MIT licensed. A smaller set of Enterprise Edition (EE) features, kept in ee/-prefixed directories, are under a separate commercial license defined in the repository's own ee/LICENSE file. GitHub's repository metadata reports the overall license as "Other" rather than a plain MIT badge because of this mixed structure β read the LICENSE and ee/LICENSE files yourself if a license determination matters for your use case.
How do I self-host Langfuse?
Per Langfuse's own documentation, self-hosting options include Docker Compose for local or quick setup (roughly a 5-minute setup per the docs), a single-VM deployment, Kubernetes via a Helm chart, and infrastructure-as-code templates for AWS, Azure, and GCP. See langfuse.com/docs for current instructions.
What does Langfuse Cloud cost?
Langfuse Cloud has four tiers: Hobby (free, 50k units/month, 30 days retention), Core ($29/month, 100k units/month included), Pro ($199/month, 3 years retention, SOC2/ISO 27001/HIPAA available), and Enterprise ($2,499/month, uptime and support SLA). These prices apply only to the managed Cloud product, not self-hosting.
Who owns or maintains Langfuse now?
As of January 2026, Langfuse is part of ClickHouse, Inc., per the copyright header in the GitHub repository's own LICENSE file ("Copyright (c) 2023-2026 ClickHouse, Inc."). Langfuse was founded in 2023 and went through Y Combinator's Winter 2023 (W23) batch.
What can I trace and evaluate with Langfuse?
Langfuse traces complex logs, chains, and user sessions; supports LLM-as-a-judge scoring, code-based evaluators, and user feedback collection; provides datasets for continuous evaluation; and includes an in-UI Playground for testing prompts, per its own documentation.
Does Langfuse run LLM inference itself?
No. Langfuse observes, traces, and evaluates calls your application makes to LLM providers such as OpenAI β it does not run model inference itself. The site's own directory lists Langfuse's engine type as "external" for this reason.
What frameworks does Langfuse integrate with?
Langfuse integrates with the OpenAI API, LangChain, LlamaIndex, and other frameworks, per its own documentation, alongside official Python and JavaScript/TypeScript SDKs and an OpenAPI-specified API.
How does Langfuse compare to Helicone?
Both are open-source LLM observability tools that are often compared. PromptQuorum has not yet published a dedicated Helicone review, so verify Helicone's current self-hosting, license, pricing, and feature details directly on helicone.ai β see the comparison table above for what this review could verify about Langfuse specifically.