Langfuse

Langfuse is an open-source platform for tracing and improving AI applications, built for teams that need to connect agent behavior with prompts and evaluations. It is strongest when a team wants those workflows together and needs the option to self-host. Its main trade-off is that useful retention and access controls sit behind paid plans.

Overview

Langfuse brings observability, prompt management, and evaluation into one development workflow. Traces record prompts, model responses, token usage, latency, tool calls, retrieval steps, and metadata, giving teams a practical way to inspect how an application behaved. Cost and token tracking are also supported.

The core is MIT-licensed and can run on customer infrastructure using Docker, including Docker Compose, Kubernetes, AWS, and Azure. Langfuse Cloud encrypts traffic with TLS 1.2+ and stored data with AES-256. Langfuse reports annual SOC 2 Type 2 and ISO 27001 audits, external penetration tests, GDPR compliance, a DPA, and a HIPAA-ready region. Role-based access control is supported, with enterprise single sign-on through OIDC.

Key features

Tracing, prompts, and evaluation

Tracing covers model calls, tools, and retrieval, making Langfuse a better fit for teams investigating multi-step agent behavior than for those concerned only with standalone model requests. Prompt Management stores and versions prompts, supports deployment and testing, and links prompts to traces; that connection helps teams relate a change in prompt to observed application behavior.

Evaluation combines LLM-as-a-judge and code evaluators with user feedback, manual labeling, datasets, and experiments. This mix supports both automated checks and human review, though teams will need to decide which evaluation methods suit their own applications.

Integration and deployment

Python and JS/TS SDKs, OpenTelemetry, an API, CLI, MCP Server, and integrations with popular AI frameworks give engineering teams several ways to connect their applications. Cloud and self-hosted deployment make it possible to choose between a managed service and operating the MIT-licensed core on customer infrastructure. Self-hosting is attractive for teams that want deployment control, but it also means running the service themselves.

Pricing

Hobby is free at no cost and includes 50k units per month, 30 days of data access, two users, and community support via GitHub. It is a sensible starting point for a small team, but the user cap and month-long access window limit its usefulness for shared, ongoing production analysis.

Core costs $29 a month, billed monthly. It includes 100k units per month, charges $8/100k units for additional usage, retains data for 90 days, allows unlimited users, and includes in-app support. This is the most direct upgrade for teams that need broader collaboration and longer access without a large monthly commitment.

Pro costs $199 a month, billed monthly, and includes the same 100k-unit allowance and $8/100k-unit overage rate, plus three years of data access, unlimited annotation queues, high rate limits, and prioritized in-app support. The higher price is most defensible for teams that rely on long-running analysis, annotation, or greater throughput.

The Teams Add-on costs $300 a month, billed monthly, and adds Enterprise SSO, SSO enforcement, fine-grained RBAC, and a dedicated Slack or MS Teams support channel. It is aimed at organizations with stricter identity and access requirements, rather than teams seeking only more usage.

Enterprise costs $2,499 a month, billed monthly. It includes 100k units per month, additional usage at $8/100k units, audit logs, a SCIM API, custom rate limits, uptime and support SLAs, and a named lead support engineer. Those commitments make it an enterprise operations option, not a cost-effective default for small teams.

Platforms

Langfuse is available through an API, as a web service, and as self-hosted software. Cloud and customer-managed deployment give teams a choice between managed operation and control over their own infrastructure.

Who it's for

Langfuse suits AI application teams that need to trace agent activity, manage prompts, and evaluate results in a shared workflow. Its self-hosted option and reported compliance measures may also matter to organizations with deployment or governance requirements. It is less compelling for an individual or small team that can work within Hobby's two-user and 30-day access limits, or for a team that needs only a narrow tracing tool.

Pros and cons

• Pros: Traces connect prompts, model responses, tools, retrieval, latency, and usage, giving teams a broad view of application behavior.
• Pros: Prompt versioning and varied evaluation methods keep improvement workflows close to operational traces.
• Pros: The MIT-licensed core can be self-hosted, while Cloud offers encryption and reported compliance measures.
• Cons: Hobby limits teams to two users and 30 days of data access, which can constrain collaboration and historical review.
• Cons: Usage beyond the included allowance costs extra on paid plans, so the monthly starting price is not the full cost for higher-volume teams.
• Cons: SSO and fine-grained RBAC require the $300 a month Teams Add-on, while key enterprise commitments require the $2,499 a month plan.

Alternatives

Consider Opik if you want a freemium option with a free trial and an open-source observability and evaluation core that can run locally. LangWatch is another freemium, open-source choice; its Developer plan is free forever and includes 50k events per month with 14-day data access. Choose W&B Weave if its Free plan's 1 GB monthly Weave data ingestion, 5 GB monthly storage, and evaluations, tracing, and scorers match your needs.

OpenLIT is worth considering when unlimited self-hosted usage, users, and projects on a free plan are the priority. Arize AX offers a free SaaS plan with 25k trace spans per month, 15-day retention, and unlimited users and evaluations. Laminar may suit a smaller free deployment, with 1 GB of data, one project, one seat, and seven-day retention.

Traceloop is an alternative for teams considering OpenLLMetry, an open-source observability project built on OpenTelemetry. Helicone is another freemium, open-source option with a free trial.

For more options, browse AI Agent Observability Tools, Prompt Management Software, LLM Observability Tools, Model Monitoring Software, AI LLM Observability Tools, and AI Prompt Management Software.

Verdict

Choose Langfuse if your AI application team wants traces, prompt workflows, and evaluations in one platform, with the flexibility to self-host. Its integrated development loop is the main reason to choose it; teams that need long retention, SSO, or stronger service commitments should budget for higher plans, while very small teams may find Hobby's caps restrictive.