--- title: Langfuse vs. Datadog for LLM Observability & Agent Tracing description: Langfuse is the open-source alternative to Datadog Agent Observability. September 2026 comparison of self-hosting, unit vs LLM-span pricing, and evals. tags: [comparison] --- # Langfuse vs. Datadog This guide outlines the key differences between **[Langfuse](/)** and **[Datadog Agent Observability](https://www.datadoghq.com/products/ai/agent-observability/)**. All Datadog facts were checked against public Datadog sources in September 2026. ## TL;DR **Choose Langfuse** if you want: **1. MIT-licensed self-hosting.** Cloud and self-hosted run the same product. Tracing, evaluations, prompt management, and experiments are in the free OSS distribution. Used by 21 of the Fortune 50. **2. Months to years of queryable history.** Core is $29/month with 90 days. Pro includes 3 years on ClickHouse. **3. Purpose-built AI quality workflows.** Code evaluators on live observations, experiments in the UI and CI, and prompt-version metrics. **Choose Datadog Agent Observability** if you want APM and LLM tracing in one Datadog deployment. You accept SaaS-only hosting and 15-day default trace retention. **Many teams run both:** Datadog for APM and infrastructure, Langfuse for AI engineering, from the same OpenTelemetry instrumentation. See [Using Langfuse and Datadog together](#coexistence). --- ## Open source and distribution [#open-source-and-distribution] Langfuse is open source (MIT). [Self-hosting](/self-hosting) is a first-class deployment mode and uses the same codebase as Langfuse Cloud. Datadog Agent Observability is proprietary SaaS. Traces are stored on [Datadog-hosted sites](https://docs.datadoghq.com/getting_started/site/). The product is [unavailable](https://docs.datadoghq.com/llm_observability/) on Datadog's government sites (US1-FED, US2-FED). | | Langfuse | Datadog Agent Observability | | -------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- | | License | **[MIT](https://github.com/langfuse/langfuse)** | Proprietary SaaS | | GitHub stars | [![Langfuse GitHub stars](https://img.shields.io/github/stars/langfuse/langfuse?style=for-the-badge&label=%20&labelColor=black&color=orange)](https://github.com/langfuse/langfuse) | N/A (platform not on GitHub) | | Self-hosting | **[Every tier, including free](/pricing-self-host)** | Not offered. Traces are stored on Datadog-hosted sites | | Air-gapped operation | Supported | Not offered | | Data residency | Cloud EU, US, Japan; [HIPAA region](/security/hipaa); any region via self-hosting | Nine hosted sites (US, EU, Japan, Australia, UK, US-gov); Agent Observability is unavailable on the two gov sites | | Published Cloud Enterprise | **[$2,499/month](/pricing)** | Custom | --- ## History and data plane [#history-and-data-plane] Langfuse writes each LLM call, tool execution, and agent step to open-source [ClickHouse](https://clickhouse.com/). Cloud and self-hosted share the same engine as of [Langfuse v4](/docs/v4). High-volume reads use the [Observations](/docs/api-and-data-platform/features/observations-api) and [Metrics](/docs/metrics/features/metrics-api) APIs. Self-hosters can query ClickHouse in SQL. Datadog Agent Observability retains traces and spans for **15 days** on Free, Pro, month-to-month, and annual plans. Add-ons extend traces to 30, 60, or 90 days and experiments to 6, 9, or 12 months. Datasets are versioned separately and [kept for 3 years](https://www.datadoghq.com/products/ai/agent-observability/). There is no published 365-day tier for Agent Observability traces. | | Langfuse | Datadog Agent Observability | | ------------------------ | ------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------- | | Storage engine | **ClickHouse** (Apache 2.0) | Datadog-operated SaaS store | | Default trace retention | 30 days Hobby, **90 days Core**, **3 years** [Pro and Enterprise](/pricing) | **15 days** on Free, Pro, M2M, and annual | | Extended trace retention | Included in the plan window; [project retention policies](/docs/administration/data-retention) on Pro+ | Paid add-on to 30, 60, or 90 days. Rates in [Pricing](#pricing) | | Experiment retention | Same window as traces | 15 days on-demand; 90 days on committed plans; add-ons to 6 / 9 / 12 months | | Dataset retention | Same project data plane | 3 years, versioned separately | --- ## Pricing [#pricing] Langfuse bills [units](/docs/administration/billable-units): one unit is a trace, an observation, or a score. Datadog Agent Observability bills LLM inference spans (one call to an LLM provider). Tool, workflow, agent, embedding, and retrieval spans are free on Datadog's [published pricing](https://www.datadoghq.com/pricing/list/). Span mix changes the crossover; the [editable pricing model](/pricing-comparison-sheet) takes it as an input so you can find yours. | | Langfuse | Datadog Agent Observability | | ----------- | -------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | | Free tier | 50,000 units/month, 2 users, 30 days | 40,000 LLM spans/month, 15-day retention, full feature access | | Paid entry | **Core $29/month** (100,000 units, unlimited users) | **Pro $160/month billed annually** ($200 month-to-month) for the first 100,000 LLM spans | | Usage rate | $8/100k units after the included 100k, graduating to $6/100k at 50M+ | $3.50 per 10k additional LLM spans (annual); $4.20 M2M; $5.00 on-demand | | Retention | 90 days Core, **3 years** Pro and Enterprise | 15 days included; 30 / 60 / 90 days at $1.50 / $3.00 / $4.00 per 10k LLM spans | | Seats | [None on Core and above](/pricing) | None published for Agent Observability | | Evaluations | Scores bill as units | Eval LLM calls bill as LLM spans; no separate eval product fee | | Enterprise | [$2,499/month](/pricing) | Custom | ### Worked example [#worked-example] 500,000 traces/month · 10 observations per trace · 2 LLM calls per trace · 1 score per trace · 5 users. | | Base | Usage | Retention | **Total/month** | | ----------------------------------- | ------- | ------- | --------- | --------------- | | **Langfuse Pro (3-year retention)** | $199.00 | $422.00 | $0.00 | **$621.00** | | **Datadog Pro (90-day retention)** | $160.00 | $315.00 | $400.00 | **$875.00** | Public annual list prices. Edit the inputs yourself: [Langfuse vs Datadog pricing model](/pricing-comparison-sheet). Langfuse: 6,000,000 units, 100k included → $199 base + $72 (100k–1M at $8/100k) + $350 (1M–6M at $7/100k) = **$621**, 3-year history included. Datadog: 1,000,000 LLM spans → $160 base + 90 × $3.50 ($315) + 90-day retention add-on 100 × $4.00 ($400) = **$875**. Month-to-month Pro is $200 + $4.20/10k. [See pricing](/pricing) · [Start Free](https://cloud.langfuse.com) · [Talk to us](/talk-to-us) --- ## Tracing and stack context [#tracing] Both platforms capture hierarchical traces of LLM applications and agents, including token usage, latency, errors, and cost. Datadog Agent Observability correlates an LLM span with APM, infrastructure, logs, and RUM in the same Datadog account. Those signals come from the rest of Datadog. Langfuse records a [trace](/docs/observability/data-model) per request, with typed [observations](/docs/observability/features/observation-types) for LLM calls, tools, and retrieval. Multi-turn chats group into [sessions](/docs/observability/features/sessions). Agent runs render as [agent graphs](/docs/observability/features/agent-graphs). [Cost](/docs/observability/features/token-and-cost-tracking) is attributed on each generation. | Feature | Langfuse | Datadog Agent Observability | | -------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- | | SDKs | [Python and JS/TS](/docs/observability/sdk/overview) on OpenTelemetry; other languages via the [OTLP endpoint](/integrations/native/opentelemetry) | [Python, Node.js, and Java](https://docs.datadoghq.com/llm_observability/instrument/sdk/); HTTP API for other languages | | Auto-instrumentation | [100+ integrations](/integrations), including OpenAI, LangChain, LlamaIndex, Vercel AI SDK, LiteLLM, CrewAI | [Supported LLM providers and agent frameworks](https://docs.datadoghq.com/llm_observability/instrument/auto_instrumentation/) | | OpenTelemetry | OTel-native SDKs + OTLP endpoint | [OTLP intake](https://docs.datadoghq.com/llm_observability/instrument/otel_instrumentation/) for GenAI semantic conventions 1.37+ and OpenInference | | Platform context | Sessions, users, environments, releases, agent graphs | Correlation with Datadog APM, infrastructure, logs, and RUM | | Production insights | [Dashboards](/docs/metrics/features/custom-dashboards), [Pulse](/docs/observability/features/pulse), filter search | [Patterns](https://docs.datadoghq.com/llm_observability/investigate/patterns/) (topic clustering of production traffic) | | Sensitive data | [SDK and OTel masking](/docs/observability/features/masking) | Sensitive Data Scanner included with Agent Observability usage | --- ## Evaluation and experiments [#evaluation-and-experiments] [Langfuse evaluations](/docs/evaluation/overview) include [LLM-as-a-judge evaluators](/docs/evaluation/evaluation-methods/llm-as-a-judge), [code evaluators](/docs/evaluation/evaluation-methods/code-evaluators) on live observations and experiments, custom scores via SDK/API, [annotation queues](/docs/evaluation/evaluation-methods/annotation-queues), and dataset experiments on Cloud and self-hosted. [CI/CD gates](/docs/evaluation/experiments/experiments-ci-cd) fail a pipeline on experiment results. Datadog provides [LLM-as-a-judge evaluators](https://docs.datadoghq.com/llm_observability/investigate/evaluations/) from a template or your own prompt, annotation queues, end-user feedback, and external evaluations via API. [Experiments](https://docs.datadoghq.com/llm_observability/improve/experiments/setup/) are started from the Python SDK with function- or class-based evaluators; datasets and run comparison live in the UI. Eval LLM calls bill as LLM spans. | Feature | Langfuse | Datadog Agent Observability | | ------------------------------- | -------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------ | | LLM-as-a-judge | [Yes](/docs/evaluation/evaluation-methods/llm-as-a-judge) (observations + experiments) | Templates or your own prompt | | Deterministic online code evals | [Yes](/docs/evaluation/evaluation-methods/code-evaluators), native | External via API or third-party frameworks; function evaluators in experiments | | Experiments | [UI + SDK](/docs/evaluation/experiments/experiments-via-sdk), versioned datasets | Python SDK datasets and experiments | | CI/CD gates | [GitHub Action](/docs/evaluation/experiments/experiments-ci-cd) | Via SDK / custom | | Annotation queues | [UI + API](/docs/evaluation/evaluation-methods/annotation-queues) | Yes | | Where evals run | Cloud or self-hosted | Datadog SaaS | --- ## Prompt management [#prompt-management] Edit the prompt in the UI. The running app fetches it. In Langfuse, pass the fetched prompt into the generation to [link it](/docs/prompt-management/features/link-to-traces), so each version gets cost, latency, and scores. Datadog [Prompt Tracking](https://docs.datadoghq.com/llm_observability/instrument/prompt_tracking/) records which prompt a span used and does not serve it. | | Langfuse | Datadog Agent Observability | | --------------------------- | ---------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | Who fetches | [Python and JS/TS](/docs/prompt-management/get-started), then [cached](/docs/prompt-management/features/caching) | Python (`ddtrace>=4.13`), [`LLMObs.get_prompt()`](https://docs.datadoghq.com/llm_observability/configure/prompt_management/) | | How you aim a version | [Labels](/docs/prompt-management/features/prompt-version-control) (`production`, `staging`, custom) | `DD_ENV`, numeric `version`, Feature Flags targeting | | One prompt includes another | [Prompt references](/docs/prompt-management/features/composability) | Concatenate in application code | | Where it runs | Cloud or self-host | Datadog SaaS | --- ## Using Langfuse and Datadog together [#coexistence] Keep Datadog for APM and infrastructure and add Langfuse for AI engineering. The [Python SDK](/docs/observability/sdk/overview) sets up OpenTelemetry when you initialize the client. The default [filter](/docs/observability/sdk/advanced-features#filtering-by-instrumentation-scope) exports Langfuse SDK spans, `gen_ai.*` attributes, and known LLM instrumentors. ```python from langfuse import Langfuse langfuse = Langfuse() ``` To keep Langfuse spans out of Datadog, pass an [isolated TracerProvider](/docs/observability/sdk/advanced-features#isolated-tracer-provider). JS/TS still registers a `LangfuseSpanProcessor` on your `NodeSDK`. A collector can also fan out one export to both backends. --- ## Which should you choose [#which-should-you-choose] **Langfuse** when you want MIT self-hosting, years of queryable history, or purpose-built [AI engineering workflows](/academy/ai-engineering-loop). **Datadog Agent Observability** for APM and LLM tracing in one Datadog deployment. **Both** when SRE keeps Datadog and the AI team uses Langfuse. See [Using Langfuse and Datadog together](#coexistence). --- ## Why teams move away from Datadog Agent Observability [#why-teams-move-away] - **SaaS only.** Datadog Agent Observability [cannot be self-hosted](#open-source-and-distribution) and is unavailable on Datadog gov sites. - **Quality loop is Python-SDK-first.** Datadog documents prompt fetch and experiment runs for Python. Online code evals go through the API or another framework. Langfuse runs [code evaluators and UI and CI experiments](#evaluation-and-experiments) on Cloud and self-host. - **15-day default traces.** Datadog Agent Observability includes 15 days. A 90-day window is a paid add-on on the AI SKU, with no published year-long tier. In the [worked example](#worked-example), that is $875/month vs $621 on Langfuse Pro with 3-year history. --- ## Why teams choose Langfuse [#customers] - **[Canva](/users/canva)** built a multi-agent support system for a product with **250 million** monthly active users. Help Assistant is Java via OpenTelemetry; Omni Agent is Python via the Langfuse SDK. LLM-as-a-judge evaluators score both systems across **15–20** metrics, and domain experts build evaluators without engineering help. They self-hosted first, then moved to Cloud. - **[SumUp](/users/sumup)** rolled AI support to **35+** markets and deflected almost **50%** of conversations, cutting external BPO cost by **30%**. They started self-hosted for the PoC, then moved to Langfuse Cloud. - **[Merck](/users/merckgroup)** runs about **80** of 300+ GenAI use cases on Langfuse with **200+** people building on it. They self-host for data sovereignty and provision via API. --- ## Switching from Datadog Agent Observability [#switching-from-datadog] Move live instrumentation and durable assets. Trace history usually stays in Datadog. Most teams see first traces in Langfuse the **same day**. - **Instrumentation:** initialize the [Langfuse SDK](/docs/observability/sdk/overview) or send OpenTelemetry to the [OTLP endpoint](/integrations/native/opentelemetry). - **Datasets, prompts, and evaluators:** import datasets via [CSV, SDK, or API](/docs/evaluation/experiments/datasets). Recreate prompts and judges in [prompt management](/docs/prompt-management/get-started) and [evaluators](/docs/evaluation/overview). - **Trace history:** stays in Datadog for its retention window. [Talk to us](/talk-to-us) if you want help planning a migration. --- ## Start free: Cloud or self-host [#get-started] [Langfuse Cloud](https://cloud.langfuse.com) Hobby includes 50k units/month, no credit card. Or [self-host](/self-hosting) the MIT stack. Explore the [example project](/docs/demo) first. Already on Datadog Agent Observability? See [Switching from Datadog Agent Observability](#switching-from-datadog). - [Get started free](https://cloud.langfuse.com) - [Talk to us](/talk-to-us) --- ## FAQ [#faq] ### Is Langfuse an alternative to Datadog Agent Observability? [#langfuse-datadog-alternative] Yes. Langfuse is an open-source alternative to Datadog's AI observability SKU. See [Which should you choose](#which-should-you-choose). ### Can I use Langfuse and Datadog at the same time? [#langfuse-and-datadog-together] Yes. Keep Datadog APM and add Langfuse for LLM traces and the AI quality loop. See [Using Langfuse and Datadog together](#coexistence). ### How does Datadog Agent Observability pricing compare to Langfuse? [#datadog-llm-observability-pricing] Datadog Agent Observability bills LLM inference spans (Pro **$160/month** billed annually for 100k spans, September 2026). Langfuse bills units from **$29/month**. On the rates in our [editable public model](/pricing-comparison-sheet), 500k traces / 10 observations / 2 LLM calls / 1 score per trace is **$621** Langfuse Pro with 3-year history vs **$875** Datadog Pro with the 90-day retention add-on. Details in the [worked example](#worked-example). ### Can I self-host Datadog Agent Observability? [#self-host-datadog] No. Agent Observability is Datadog-hosted SaaS. Langfuse [self-hosts](/self-hosting) on every tier. ### Did joining ClickHouse change the Langfuse product? [#clickhouse-acquisition] No. Langfuse already ran on ClickHouse. The [announcement](/blog/joining-clickhouse) states the MIT license, self-hosting, Cloud endpoints, and roadmap stay the same, with more capacity to ship. Cloud Core and Pro remain self-serve; OSS self-host has no sales motion. Details in [clarifications](/resources/engineering/clarifications#clickhouse). ### How do I migrate from Datadog Agent Observability to Langfuse? [#migrate-from-datadog] Point live traffic at Langfuse and recreate datasets, prompts, and evaluators. History usually stays in Datadog. See [Switching from Datadog Agent Observability](#switching-from-datadog). This comparison is out of date? Please [raise a pull request](https://github.com/langfuse/langfuse-docs/tree/main/content/resources/engineering) with up-to-date information. --- ## Agent Instructions This page is part of the [Langfuse](https://langfuse.com) documentation, published as plain Markdown for AI agents. Every page is available as Markdown by appending `.md` to its URL, or by sending an `Accept: text/markdown` header. This page: `https://langfuse.com/resources/engineering/langfuse-vs-datadog.md`. ### Querying these docs If the answer is not on this page, query the documentation instead of guessing: - **Semantic search** across all Langfuse docs, returning an answer with the relevant pages and excerpts. Ask a specific, self-contained question: ```bash curl -sG "https://langfuse.com/api/search-docs" --data-urlencode "query=How do I trace a LangGraph agent?" ``` - **Index of every page**: , with per-section indexes [llms-docs.txt](https://langfuse.com/llms-docs.txt), [llms-integrations.txt](https://langfuse.com/llms-integrations.txt), and [llms-self-hosting.txt](https://langfuse.com/llms-self-hosting.txt). ### Before writing Langfuse code - **Install the [Langfuse Agent Skill](https://langfuse.com/docs/api-and-data-platform/features/agent-skill).** It encodes Langfuse's own best practices for instrumentation, prompt management, and evaluation, and materially improves results. - **Read [What does a good trace look like?](https://langfuse.com/docs/observability/best-practices.md)** before instrumenting an application. - **Verify endpoints, parameters, and response fields** against the [API reference](https://api.reference.langfuse.com) instead of inferring them from code examples. - **Use the [Langfuse CLI](https://langfuse.com/docs/api-and-data-platform/features/cli)** (`npx langfuse-cli api `) to read or write traces, prompts, datasets, and scores from the terminal. Found an error in these docs? Please open an issue at .