Langfuse

Langfuse is an open-source AI engineering platform for teams building and refining agent applications. It suits engineers who need observability, prompt lifecycle controls, and evaluation in the same workflow, particularly when self-hosting or compliance controls matter. The main trade-off is that longer retention, larger teams, and advanced access features require paid plans.

Overview

Langfuse connects production behavior with the material used to improve it. Tracing records prompts, model responses, token usage, latency, tool calls, retrieval steps, and metadata, so teams can investigate multi-step agent runs rather than viewing only isolated model requests. Cost and token tracking add a direct way to monitor spend.

The MIT-licensed core can run on customer infrastructure through Docker, Docker Compose, Kubernetes, AWS, or Azure. Langfuse Cloud uses TLS 1.2 or newer for traffic and AES-256 for stored data. The company 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 included, while enterprise single sign-on uses OIDC.

Key features

Tracing and evaluation

Agent, LLM, and retrieval tracing covers the calls and intermediate steps that commonly make debugging difficult. Evaluation combines LLM-as-a-judge and code evaluators with user feedback, manual labeling, datasets, and experiments. That mix supports automated regression checks alongside human review, making the platform useful for teams iterating on quality rather than merely collecting logs.

Prompt workflows and integrations

Prompt Management stores, versions, deploys, and tests prompts, then links them to traces. This gives developers a way to relate a prompt change to observed application behavior. Python and JavaScript/TypeScript SDKs, OpenTelemetry, an API, CLI, MCP Server, and integrations with popular AI frameworks provide several integration routes. Self-hosting is the stronger choice for organizations that need infrastructure control; Cloud is more suitable when operating that stack is undesirable.

Pricing

Hobby — free. It includes 50,000 units per month, 30 days of data access, two users, and community support through GitHub. The short retention and strict seat limit make it a sensible sandbox or small proof of concept, not a durable shared workspace.

Core — $29 a month, billed monthly. The plan includes 100,000 units monthly, 90 days of data access, unlimited users, and in-app support. Extra usage costs $8 per 100,000 units. It is the practical starting point for a growing team, although retention remains limited and advanced controls are absent.

Pro — $199 a month, billed monthly. It keeps the 100,000-unit allowance and $8-per-100,000-unit overage, extends data access to three years, adds unlimited annotation queues and high rate limits, and provides prioritized in-app support. Choose it when historical analysis and heavier evaluation workflows justify the substantial increase over Core.

Teams Add-on — $300 a month, billed monthly. This adds enterprise SSO, SSO enforcement, fine-grained RBAC, and a dedicated Slack or Microsoft Teams support channel. It addresses identity and collaboration requirements rather than increasing observability capacity.

Enterprise — $2,499 a month, billed monthly. Enterprise includes 100,000 units monthly, $8 per additional 100,000 units, audit logs, a SCIM API, custom rate limits, an uptime SLA, a support SLA, and a named lead support engineer. It is aimed at organizations that need contractual reliability and centralized administration.

Platforms

Langfuse is offered through its web application and API, with deployment both as hosted Cloud and as self-hosted software. The self-hosted route supports Docker-based deployments on customer infrastructure.

Who it's for

Choose Langfuse for an engineering team operating AI agents that needs traces, prompt versioning, and evaluations tied together. It is especially compelling for regulated or infrastructure-conscious organizations because the core is open source and deployment can remain on your systems. Hobby works for early experimentation; Core fits shared development; Pro and Enterprise suit teams needing longer history, higher limits, stronger support, or governance. It is less suitable if you need only a lightweight prompt library or have no use for application tracing.

Pros and cons

• Pro: One workflow links agent traces, prompt changes, and evaluation results.
• Pro: Open-source MIT licensing and multiple self-hosting targets support infrastructure control.
• Pro: Security and compliance measures, including reported SOC 2 Type 2 and ISO 27001 audits, support demanding deployments.
• Con: Hobby keeps data for only 30 days and permits two users.
• Con: Three-year retention, high rate limits, and unlimited annotation queues start at the much higher-priced Pro tier.
• Con: Enterprise identity, audit, SLA, and provisioning features require the Teams Add-on or Enterprise.

Alternatives

For a prompt-focused catalog, compare Vellum, which offers a free entry plan with included tasks and pay-as-you-go credits across web and many client platforms. PromptHelm is a fit when a gateway-oriented free tier with 10,000 calls monthly, prompt/version caps, one seat, and seven-day logs is enough. Agenta suits small teams wanting a free plan with two members, 5,000 agent runs, 20 evaluations, and one-week trace retention, or its open-source option. Impromptо is an alternative for teams seeking a low-cost playground plan with team workspaces. Modular Prompt Library targets personal prompt organization with free workspaces and testing using your own API keys. Prompt Assay, PromptDocker, and PromptFluent are other prompt-management alternatives with their own free or trial options.

Browse Prompt Management Software, AI Prompt Management Software, API Analytics Software, AI LLM Evaluation Tools, AI LLM Observability Tools, and AI Agent Observability Tools for broader category choices.

Verdict

Langfuse is a strong choice for teams that want prompt operations, agent observability, and evaluation in one extensible platform, with self-hosting available when control is essential. Look elsewhere if your needs stop at prompt organization or if the retention, seats, and governance required for production make its paid tiers disproportionate.