Research

Langfuse Alternatives: Who Should Stay, and Where a Prompt Test Fits

Last checked 18 November 2026.

Langfuse is an open-source platform for tracing, prompt management and evaluation of LLM apps. In January 2026 it joined ClickHouse, which prompted some teams to check what changes. This guide covers what we found, what Langfuse does better than a browser tool, who should stay with it, where Spring Prompt fits, and how to bring a dataset across.

Have a Langfuse dataset? Load the export into a Prompt test and read every model's answers side by side. Test your prompt →

What's changed

On 16 January 2026, Langfuse announced: “We're excited to announce that Langfuse is joining ClickHouse.” Its announcement said Langfuse Cloud “keeps running as‑is”, that it “stays open source and self‑hostable”, and, asked whether it's still open source: “Yes. No licensing changes planned.” ClickHouse's post says Langfuse “remains 100% open-source under its existing MIT license for core features”.

Ten months on, the repository's README says “Since January 2026 we're part of ClickHouse”, and the licence line reads “This repository is MIT licensed, except for the ee folders.”

Sources: Langfuse's announcement on GitHub, dated 16 January 2026; ClickHouse's post; the Langfuse repository and Langfuse pricing, checked 18 November 2026.

Langfuse Cloud plans

PlanPriceIncluded
HobbyFree50,000 units a month, 30 days of data access, 2 users
Core$29 a month100,000 units a month, 90 days of data access, unlimited users
Pro$199 a month100,000 units a month, 3 years of data access
Enterprise$2,499 a month100,000 units a month, 3 years of data access

Extra usage on paid plans starts at $8 per 100,000 units and falls with volume. Self-hosting is free, and Langfuse has a separate page for self-hosted pricing.

What Langfuse does better

  • Tracing. Langfuse traces LLM calls, retrieval, embeddings and agent actions, built on OpenTelemetry. Spring Prompt doesn't trace. Its Live package samples real calls to build tests, not to monitor production.
  • Prompt management. Versioned prompts in one place, with a playground for trying them.
  • Evaluation at scale. LLM-as-a-judge, code evaluators, user feedback and manual labelling, with datasets and experiments that run over every item. A Prompt test is capped at 24 cases.
  • Self-hosting. Docker Compose for one machine, Helm for Kubernetes, and Terraform templates for AWS, Azure and GCP. If your data can't leave your infrastructure, this matters. Spring Prompt is hosted only.
  • Dashboards. Latency, cost and quality over time, with a metrics API and alerts.

Who should stay with Langfuse

If you want production traces, prompt versions and evals in one tool, need to self-host, or already have datasets and evaluators set up there, Langfuse covers more than we do, and nothing we found in the ClickHouse deal is a reason to leave. A Prompt test is useful alongside it for one job: comparing models on a prompt quickly, with every answer and its cost on one page.

Where Spring Prompt fits

  • Prompt test. Paste a prompt in a browser. We draft test cases and checks for you to edit, run the models you pick side by side, check every answer against your rules, and show the cost and time of each answer next to whether it passed. Nothing to deploy.
  • Live. A two-line package, springprompt on PyPI and npm, that copies a sample of your app's real model calls so you can test them on other models. See the Live docs.
  • Check before merge. incremento/springprompt-action@v1 runs a changed prompt or model on a saved test and returns pass, review or fail. See how it works.
  • Independent benchmarks. Our benchmarks score current models on business tasks and show their real answers.

Langfuse vs Spring Prompt

LangfuseSpring Prompt
Main jobTracing, prompt management and evalsTesting a prompt across models
SetupSDK or OpenTelemetry in your appPaste a prompt; optional two-line Live package
Writing testsDatasets and evaluators you buildWe draft cases and checks from your prompt; you edit them
Test sizeWhole datasetsUp to 24 cases and 10 checks
Production tracesYesNo; Live samples calls for testing
Licence and hostingMIT core (except ee folders); cloud or self-hostedHosted only
PricingHobby free; Core $29, Pro $199, Enterprise $2,499 a monthPay for the tests you run; free trial credit

Get your data out of Langfuse

  • Datasets. Langfuse added CSV and JSON export of datasets from the UI in June 2025. Click the export button at the top right of the table; the export is queued, and the file appears under Project Settings › Exports. Filters on the table apply to the export, and every column is included.
  • Through the API or SDKs. langfuse.get_dataset(...) in Python returns a dataset's items, each with input, an optional expectedOutput and metadata.
  • Prompts. Copy out the text of each prompt version you run in production.

What loads into a Prompt test

In a new Prompt test, choose Import from my app, then paste or upload the file. A Langfuse dataset loads with no model call, whether it's the UI's JSON or CSV export or the API's list of items:

Item's inputWhat comes in
Named fields ({"question": ...})Each field as a {{variable}}, one case per item
A chat (a list of messages)The conversation as the input, with the system message kept
Plain textThe text as the input

Each item's expectedOutput becomes the case's expected answer, with a judged check that the answer matches it. Archived items and metadata are left out, and the import says so. A dataset holds inputs, not your prompt, so the import makes a stand-in from the fields; paste your real prompt over it, keeping the {{variables}}.

A Prompt test takes up to 24 cases and 10 checks, so a large dataset is cut to the first ones. Choose the items that went wrong or nearly did. Evaluators written as code don't carry over; describe what each checked and we can draft a replacement. API keys in the file are removed before anything is stored.

Other guides: LangSmith, Braintrust, Promptfoo and Helicone.

Bring your dataset, or start from a prompt. Run the models you're choosing between on the same inputs and read every answer side by side. Test your prompt →

Quick answers

Is Langfuse still open source after joining ClickHouse?

Yes. Langfuse announced it was joining ClickHouse on 16 January 2026 and said no licensing changes were planned. Its repository says it is MIT licensed, except for the ee folders. We checked on 18 November 2026.

Can I still self-host Langfuse?

Yes. Langfuse's README covers Docker Compose for a single machine, a Helm chart for Kubernetes (its preferred production setup), and Terraform templates for AWS, Azure and GCP. Its pricing page says you can self-host it for free.

How much does Langfuse Cloud cost?

Langfuse's pricing page lists Hobby free (50,000 units a month, 30 days of data access, 2 users), Core at $29 a month, Pro at $199 a month and Enterprise at $2,499 a month. Paid plans include 100,000 units a month, with extra usage from $8 per 100,000 units.

How do I export a Langfuse dataset?

Datasets can be exported as CSV or JSON from the UI. Exports are queued, then downloaded from Project Settings, Exports. The API and SDKs also return dataset items.

Can I import a Langfuse dataset into Spring Prompt?

Yes. The UI's JSON or CSV export, or the API's list of items, loads into a Prompt test without a model call. Each item's input becomes the case and its expected output the expected answer. A Prompt test takes up to 24 cases and 10 checks.

More research
All posts →