Both cover tracing, experiments, online evaluation, annotation queues and prompt versioning. Langfuse is the open-source option you can run yourself for free; Datadog is the hosted option that links agent traces to the APM and infrastructure data you already collect there. Hosting and your existing monitoring stack usually decide it.[1][6]
Langfuse is an open-source (MIT, apart from licensed add-ons) LLM engineering platform you can self-host for free, while Datadog Agent Observability, formerly LLM Observability, is a hosted module of the Datadog platform with no self-hosted option. ClickHouse acquired Langfuse in January 2026; Langfuse says its licence and self-hosting are unchanged. Both trace agents, run experiments, score production traffic and version prompts.
| Decision factor | Datadog | Langfuse |
|---|---|---|
| Ownership | Datadog, a public company (Nasdaq: DDOG); needs a Datadog account | Acquired by ClickHouse in January 2026; licence and self-hosting unchanged, per Langfuse |
| Licence | Proprietary SaaS; SDKs such as dd-trace-py are open source | MIT, except add-on features in the ee folders |
| Self-hosting | None documented | Free, via Docker or Kubernetes |
| OpenTelemetry | OTLP with GenAI conventions v1.37+ or OpenInference; reads Langfuse instrumentation | OTLP over HTTP; gRPC not supported |
| Code evaluators in production | Run in your app and submitted; automatic only in Experiments | Python or TypeScript evaluators attached to live observations |
| Correlation with backend | Linked to APM, infrastructure and RUM data | LLM and agent traces only |
| Free tier | Free: 40K LLM spans a month, 15-day retention | Hobby: 50k units a month, 30 days of data; self-hosting free |
| First paid tier | Pro: $160 a month billed annually, 100K LLM spans | Core: $29 a month, 100k units |
| Capability | Datadog | Langfuse | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | ✓ | ✓ | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | — | ✓ |
| Gateway or proxy in the request path | — | — | — |
| Licence and deployment | |||
| Open source | Partial | ✓ | Partial |
| Self-hosting | — | ✓ | Partial |
Langfuse publishes its platform under the MIT licence, apart from add-on features in licence-keyed ee folders, and documents a free self-hosted path on Docker or Kubernetes backed by Postgres, ClickHouse, Redis and object storage. ClickHouse acquired Langfuse in January 2026, and Langfuse stated that its roadmap and licence would not change and that self-hosting stays a first-class path. It has kept releasing since, most recently v4.64.0 on 19 October 2026.
Datadog Agent Observability, renamed from LLM Observability in June 2026, is proprietary and hosted only. It is not available on Datadog's US government sites, and Lapdog, its local trace viewer, is a development tool. In return, agent traces sit next to Datadog's APM, infrastructure and RUM data, so a failing agent step can be followed into the backend services behind it. If your services already report to Datadog, that link is the strongest reason to choose it; if traces must stay on your own servers, Langfuse is the one of the two that allows it.
Both accept OpenTelemetry, which keeps instrumentation portable. Langfuse acts as an OTLP backend over HTTP (not gRPC) and names OpenLLMetry and OpenLIT as compatible. Datadog accepts OTLP traces without its own SDK, using the OpenTelemetry GenAI conventions (v1.37 or later) or OpenInference, and its documentation says it also recognises OpenLLMetry and Langfuse instrumentation. Datadog notes a 3 to 5 minute ingestion delay for OTLP traces.
That makes it practical to run both for a period: for example, Langfuse for prompt work and evaluation, and Datadog for production monitoring next to the rest of your services. Langfuse Cloud users have a separate date to plan for: Langfuse Cloud becomes v4-only on 16 November 2026, which may require migration work for existing users.
Both run experiments on datasets, score sampled production traffic with LLM judges, queue traces for human review and version prompts with a playground. Langfuse's experiments can run against dataset items, production traces or local data, a GitHub Action fails a CI run below a score threshold, and since September 2026 evaluator backfills can score up to six months of past observations.
The clearest difference is code evaluators. Since May 2026 Langfuse lets you write Python or TypeScript evaluators in the UI and attach them to live observations or experiments. Datadog runs code evaluators automatically only in Experiments; for production traffic you run them in your own application and submit the scores through the SDK or API. On prompts, Datadog adds A/B tests of prompt versions and per-environment targeting rules; Langfuse deploys versions by label without code changes and compares metrics across versions.
For multi-step agents, Langfuse added an agent graph view in July 2026 and groups multi-turn runs into sessions. Datadog shows tools used and handoffs for OpenAI Agents SDK, LangGraph and CrewAI in Python. Langfuse Cloud runs in EU, US, JP and HIPAA regions; Datadog runs on several Datadog sites.
The two price on different units. As of 20 October 2026, Langfuse charges per unit ingested, counting traces, observations and scores: Hobby is free for 50k units a month with 30 days of data, Core is $29 a month for 100k units, then $8 per 100k, and Pro is $199 a month. Self-hosting has no subscription but adds the infrastructure you run. Datadog bills only spans that call an LLM provider: Free covers 40K a month with 15-day retention, and Pro is $160 a month billed annually for the first 100K, then $3.50 per 10K.
Because Langfuse counts every observation and score while Datadog counts only LLM calls, agents with many tool steps weigh more on Langfuse's meter. Model both with your own traffic using Datadog's pricing and Langfuse's pricing.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run, not only the final answer, and can act on what it finds through runtime policies, approval holds and a kill switch scoped to a run, an agent or a team. Langfuse evaluates after the fact and documents no run-time blocking. Datadog's module does not block either; it sells that separately as AI Guard, in limited availability. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if free self-hosting or prompt iteration is the requirement, Langfuse or Datadog is the better pick. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.