Langfuse

Langfuse is an open-source AI engineering platform for teams working on AI agent applications. Its tracing captures prompts, model responses, token usage, latency, tool calls, retrieval steps, and metadata, giving teams records to inspect as they debug and analyze applications. Prompt Management lets teams store and version prompts, deploy and test them, and connect them to traces. Evaluation tools include LLM-as-a-judge, code evaluators, user feedback, manual labeling, datasets, and experiments. Integrations include Python and JS/TS SDKs, OpenTelemetry, an API, CLI, MCP Server, and popular AI frameworks. Teams can self-host with Docker, including Docker Compose, Kubernetes, AWS, and Azure options; the core is MIT-licensed. The free Hobby plan includes 50k units per month, 30 days of data access, and two users. Core costs $29 a month, billed monthly, with 100k units per month included. Langfuse also reports security audits, external penetration tests, GDPR compliance, a DPA, and a HIPAA-ready region.

Langfuse brings tracing, prompt workflows, and evaluations together for AI agent development. The free Hobby plan offers a way to start, while paid plans add usage and support options at listed prices.

Langfuse suits teams building AI agent applications that need tracing, prompt management, and evaluation tools. Teams can use the hosted service or self-host the MIT-licensed core.

Langfuse is an open-source platform for teams building and improving AI agent applications. It suits developers who want traces, prompt workflows, and evaluation in a shared system, with the choice of hosted or self-managed deployment. Its breadth is the draw; the free plan’s 30-day data access and two-user limit make it a starting point, not a roomy team workspace.

Overview

Langfuse connects application behavior to the prompts and evaluations used to refine it. Traces capture prompts, model responses, token usage, latency, tool calls, retrieval steps, and metadata, so teams can examine more than just a model’s final output. Cost and token tracking are also supported.

The core is MIT-licensed and can be run on customer infrastructure with Docker, including Docker Compose, Kubernetes, AWS, and Azure deployment options. Langfuse Cloud protects traffic with TLS 1.2+ and stored data with AES-256 encryption. 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 spans model calls, tools, and retrieval, making Langfuse relevant to teams debugging multi-step agents as well as individual model interactions. Prompt Management stores and versions prompts, supports deployment and testing, and links prompts to traces. That connection gives teams a way to relate prompt changes to observed behavior rather than treating prompts as a separate asset.

Evaluation includes LLM-as-a-judge and code evaluators, alongside user feedback, manual labeling, datasets, and experiments. This combination supports both automated checks and human review, which is useful for teams iterating on agent quality. Python and JS/TS SDKs, OpenTelemetry, an API, CLI, MCP Server, and integrations with popular AI frameworks provide several routes into existing development workflows.

Pricing

Langfuse is freemium. Hobby costs nothing and includes 50k units per month, 30 days of data access, two users, and community support via GitHub. It is a sensible way for an individual or small project to begin, but the user cap and short retention can constrain shared debugging over time.

Core costs $29 a month, billed monthly. It raises the included allowance to 100k units per month, charges $8/100k units for additional usage, extends data access to 90 days, and provides unlimited users and in-app support. This is the clearest fit for teams that need shared access without a long retention window.

Pro costs $199 a month, billed monthly, with 100k units per month included and additional usage at $8/100k units. It adds three years of data access, unlimited annotation queues, high rate limits, and prioritized in-app support. Teams that need historical traces or more headroom for annotation have a reason to move up; the higher price is harder to justify if 90 days and standard in-app support are enough.

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 an access-control and support upgrade, not a replacement for a usage plan. Enterprise costs $2,499 a month, billed monthly, and includes 100k units per month, additional usage at $8/100k units, audit logs, SCIM API, custom rate limits, uptime and support SLAs, and a named lead support engineer. It fits organizations that need those controls and commitments; smaller teams are better served by Core or Pro.

Platforms

Langfuse is available through an API and web interface, or as a self-hosted deployment. Self-hosting gives teams infrastructure control, while Cloud offers managed service protections; the choice depends on deployment requirements and operational preference.

Who it's for

Langfuse is best for engineering teams developing AI agents that need to inspect traces, manage prompt versions, and evaluate results in a connected workflow. Its self-hosted option and security controls also suit organizations with specific deployment or access requirements. It is less compelling for teams seeking only basic prompt logging, or for projects that need long retention and multiple users but cannot move beyond the free plan.

Pros and cons

• Connected development workflow: Traces link to versioned prompts, while evaluations span automated scoring and human feedback.
• Flexible deployment: The MIT-licensed core can be self-hosted, and Cloud has stated encryption and compliance measures.
• Clear upgrade path: Paid plans expand data access, team access, rate limits, and support options.
• Free tier constraints: Two users and 30 days of data access can limit collaboration and historical investigation.
• Usage costs can grow: Core, Pro, and Enterprise each include 100k units monthly, then charge $8/100k units.

Alternatives

Evidently AI is worth considering for teams seeking an Apache 2.0 open-source framework with a free plan. Laminar offers a free tier with 1 GB of data, seven-day retention, one project, and one seat; choose it if those limits suit a small deployment. Maxim AI has a free Developer plan capped at three seats, one workspace, 10k logs per month, and three-day retention, making it an option for a small, bounded evaluation.

Memoturn is another free, self-hosted option. Opik offers an open-source core for local use. Qualdo-MQX has a Starter plan at $79 a month with up to 100 GB of data and 10 ML models. W&B Weave has a free plan with 1 GB monthly ingestion, 5 GB monthly storage, and evaluations, tracing, and scorers. Arize AX offers a free tier with 25k trace spans per month, 1 GB monthly ingestion, and 15-day retention. These alternatives may suit readers whose needs align more closely with their respective plan limits or deployment models.

Verdict

Choose Langfuse if your team wants agent tracing, prompt management, and evaluation in one workflow, especially if self-hosting matters. Its strongest case is that connected toolkit; look elsewhere if the free tier’s short retention and small seat allowance are deal-breakers and paid usage does not fit your budget.