Langfuse Alternatives: 9 LLM Observability Tools Compared

The best Langfuse alternatives split into two camps: tools you can self-host without a sales call (Arize Phoenix, Opik, MLflow and Laminar), and commercial platforms where self-hosting is an enterprise deal (LangSmith, Braintrust and Logfire) or a conversation with the vendor (Weave). Helicone sits awkwardly in between. Its code is Apache-2.0, but its future is unclear after a March 2026 acquisition by Mintlify and a later banner saying it is no longer part of Mintlify.

People look for Langfuse competitors for a few reasons:

• Ownership. Langfuse joined ClickHouse on January 16, 2026. It says there are "no planned changes to licensing," but some teams want an independent vendor.
• Ops load. Self-hosting Langfuse means running Postgres, ClickHouse, Redis and blob storage. Lighter tools exist.
• Focus. Some teams want deeper evals, a proxy gateway, or tracing that sits inside their existing ML platform.

If you're weighing Langfuse against LangSmith specifically, our Langfuse vs LangSmith head-to-head goes deeper. This post is the wider roundup, with every licence and hosting claim checked against the vendor's own repo or docs.

Quick comparison

Open source Langfuse alternatives you can self-host

These are the answer if you searched open source langfuse alternative or self hosted llm observability. You can run each one without a sales call.

Arize Phoenix

Phoenix is the closest like-for-like swap. It traces LLM apps, runs LLM-as-judge evals, versions datasets, tracks experiments, and has a playground with prompt versioning. It's "built on top of OpenTelemetry," and auto-instrumentation comes from Arize's OpenInference project.

Getting started is light: pip install arize-phoenix then phoenix serve, or a Docker image with a Helm chart for Kubernetes. That's far fewer moving parts than a Langfuse self-host.

One catch: Phoenix's main repository uses the Elastic License 2.0, which is source-available rather than an OSI-approved open-source licence. It says you may not provide the software to third parties as a hosted or managed service. The separate Phoenix Python client and OTel packages are Apache-2.0. For internal use the ELv2 limit rarely matters. Arize sells a managed product, Arize AX, that accepts the same instrumentation.

Opik by Comet

Opik is Apache-2.0, and its README is explicit that this covers "the full platform, backend included, not just a client SDK." It logs LLM calls and multi-step agent traces, runs datasets and experiments with code metrics and LLM-as-judge, and has a PyTest integration for CI.

Two features stand out against Langfuse: an Agent Optimizer SDK for automated prompt and agent tuning, and built-in guardrails you can switch on in a self-hosted setup. Docker Compose covers local use; Helm covers production. Comet also hosts Opik if you'd rather not run it. One recent note: Comet reported a security incident affecting Opik on comet.com on October 26, 2026. It says no customer data, accounts or API keys were accessed, that self-hosted Opik was not affected by the incident, and that it recommends upgrading to the latest version because the underlying flaw existed in earlier open-source releases.

MLflow

MLflow is Apache-2.0 and now pitches itself as "The Open Source AI Engineering Platform for Agents, LLMs & Models." Its GenAI features include tracing that's "fully compatible with OpenTelemetry," LLM-as-judge evaluation with pre-built and custom judges, and a prompt registry with versioning and automated optimization.

The pull is consolidation. If your team already tracks classic ML experiments in MLflow, LLM traces land in the same place. MLflow sits under the Linux Foundation, and you can self-host it or use a managed version on Databricks, AWS SageMaker, Azure Machine Learning and others.

Laminar

Laminar is Apache-2.0 and built for agents. Its OpenTelemetry-native SDK auto-traces the Vercel AI SDK, Browser Use, Stagehand, LangChain, OpenAI and Anthropic. Two features are unusual:

• Signals: you describe a behavior in plain English, Laminar checks every run for it, and alerts you in Slack.
• SQL over traces: query spans, metrics and events with SQL through its CLI or MCP server.

docker compose up -d starts a local self-hosted stack. For production, the README recommends either Laminar's managed platform or the fuller docker compose -f docker-compose-full.yml up -d.

Commercial Langfuse competitors

These suit teams that would rather pay than run infrastructure. Self-hosting any of them is an Enterprise-plan or account-rep matter, as the entries below show.

LangSmith

LangSmith is LangChain's commercial platform for tracing, evaluation, prompts and agent deployment. It works with many frameworks, not only LangChain, and accepts OpenTelemetry traces. Self-hosted LangSmith "is an add-on to the Enterprise plan." Pick it if you want one vendor to trace, test and host your agents. The full breakdown is in our Langfuse vs LangSmith comparison.

Searching langsmith alternatives open source? Langfuse itself is the usual answer, with Phoenix and Opik close behind.

Braintrust

Braintrust calls itself "the active observability platform for instrumenting, understanding, and improving agents." Its center of gravity is evaluation: experiments to compare prompts and models, an AI agent called Loop that analyzes data and builds evals, and a feature that finds recurring trace problems and suggests fixes.

Braintrust is a commercial platform, but its AutoEvals scoring library is MIT-licensed. For data residency, Braintrust's self-hosting option puts the data plane in your cloud while Braintrust runs the control plane. The docs say "Self-hosting is only available on the Enterprise plan." A BYOC option, where Braintrust operates the deployment in your cloud, also exists.

W&B Weave

Weave is Weights & Biases' LLM tracing and evaluation toolkit. The repository is Apache-2.0. You decorate functions with @weave.op, and it has drop-in integrations for the OpenAI Agents SDK, Claude Agent SDK and Google ADK.

The backend is W&B's. Per the Weave deployment docs, you can use multi-tenant cloud, dedicated cloud on AWS, GCP or Azure, or self-managed instances, where the docs say CoreWeave account representatives provide guidance on deployment options. Ownership note: CoreWeave completed its acquisition of Weights & Biases in May 2025, and the docs now live on CoreWeave's domain.

Pydantic Logfire

Logfire comes from the Pydantic team. It's "an opinionated wrapper around OpenTelemetry" that observes your whole application, not just LLM calls, so model calls show up next to the rest of your traces. The SDK is MIT and can export to any OTel backend. But the README says "The Logfire platform (the UI and backend) is closed source," and that you can self-host it by purchasing an enterprise licence.

Pick it if you're a Python team (especially on Pydantic AI) and want one tool for general and LLM observability.

Helicone: open source, but check its status first

Helicone is Apache-2.0 and takes a different approach from Langfuse. Instead of an SDK, you change your base URL and route model calls through its AI gateway, which logs them and adds routing and fallbacks across many models. Self-hosting runs on Docker Compose.

The status is the problem. On March 3, 2026, Mintlify announced it was acquiring Helicone and said Helicone "will continue operating in maintenance mode," with security updates, bug fixes and new models still shipping, while it would "support a smooth migration to another platform" for customers. Today, Helicone's own site carries a banner: "Helicone is no longer part of Mintlify. Details to come soon."

So neither source announces a shutdown, but Helicone's roadmap is unclear. If you already self-host it, the code stays Apache-2.0, and its repository was still receiving commits on October 28, 2026. If you're picking a tool for the next two years, wait for those details or choose another option from this list.

How to choose an LLM tracing tool

Most teams can narrow the list with five questions.

A few practical tips:

1. Instrument with OpenTelemetry where you can. Langfuse, Phoenix, MLflow, Laminar, Logfire and LangSmith all speak OTel. Standard spans make switching tools later a config change, not a rewrite.
2. Count the services before you self-host. One local process (phoenix serve) is a different job from a multi-database stack. Price the ops time, not just the licence.
3. Read the licence, not the label. "Open source" is used for MIT, Apache-2.0, source-available licences such as ELv2, and open SDKs with closed backends. Only the first two are OSI-approved open-source licences.
4. Check pricing pages yourself. Cloud billing ranges from per-seat to usage-based, and it changes often.

Observability follows your framework choice. If that's still open, see our roundup of LangChain alternatives.

Where Taku fits

Every tool above is for teams who build LLM apps and need to see inside them. Taku doesn't trace or evaluate anything. Taku is for the people on the other side of that work: an AI-native desktop workspace where you mirror an AI app, skill or workflow someone already got working, run it, remix it, and keep it. Taku is in Beta, and the Mac app is available now. The free app library shows what's ready to run.

FAQ

What is the best open source Langfuse alternative?

Arize Phoenix is the closest like-for-like match, and it runs locally with pip install arize-phoenix and phoenix serve, though its Elastic License 2.0 is source-available rather than OSI open source. For a permissive Apache-2.0 licence covering the full platform, look at Opik, Laminar or MLflow.

Is Langfuse still open source after the ClickHouse acquisition?

Yes. Langfuse joined ClickHouse on January 16, 2026, and said there are no planned licensing changes, self-hosting stays "a first-class path," and Langfuse Cloud keeps running as before. In the repository, the MIT licence covers everything outside the ee directories, which carry a separate enterprise licence.

What are the best self-hosted LLM observability tools?

Langfuse, Phoenix, Opik, MLflow, Laminar and Helicone can all be self-hosted for free. LangSmith, Braintrust and Logfire offer self-hosting only on enterprise plans or licences, and Weave's docs say CoreWeave account representatives guide self-managed deployments.

Is Helicone shutting down?

Neither Mintlify's announcement nor Helicone's site gives a shutdown date. Mintlify said in March 2026 that Helicone would run in maintenance mode after its acquisition, and Helicone's site now says it is no longer part of Mintlify, with details to come. Treat its long-term roadmap as uncertain.

Are there open source LangSmith alternatives?

Yes. Langfuse is the most common one, and Phoenix, Opik, MLflow and Laminar all cover tracing and evaluation with self-hosting at no licence cost.

Which LLM tracing tools support OpenTelemetry?

Langfuse, LangSmith, Arize Phoenix, MLflow, Laminar and Pydantic Logfire all build on or accept OpenTelemetry. Using OTel spans keeps you free to switch tools later.