Langfuse

Langfuse is an open-source LLM engineering platform for AI agent teams. It combines tracing, prompt management, evals, experiments, annotation, and metrics to debug quality, cost, and latency from prototype to production.

Key Highlights

• Trace agents and LLM calls with SDKs, OpenTelemetry, LiteLLM, or the API
• Manage prompt versions, releases, caching, and playground tests
• Run evals with datasets, scores, feedback, LLM-as-judge, and human review
• Track cost, latency, sessions, and users
• Self-host under the MIT license or use Cloud in US or EU regions

What Makes It Different

Langfuse connects the LLM engineering loop in one product. Traces, prompt versions, evals, experiments, and annotation queues sit together, so production issues can feed back into prompts and datasets.

It also stays open: OpenTelemetry, Python and JavaScript SDKs, Java and Go through OTel, LiteLLM logging, custom APIs, more than 100 integrations, and exports.

Features & Capabilities

Teams instrument an app, then inspect traces, chats, users, token usage, cost, and latency. From there they can manage prompt versions, fetch prompts without hard-coding them, test changes in the playground, and run experiments.

Quality tools include datasets, SDK and UI experiments, custom scores, user feedback, external eval pipelines, LLM-as-judge, and annotation queues. Production controls include batch export, PostHog and Mixpanel, webhooks, data masking, retention, SSO, SCIM, audit logs, and compliance reports depending on plan.

User Ratings and Testimonials

No public average rating is listed. Canva's AI team uses Langfuse to trace and debug generative design features, and Langfuse reports 2,300+ customers, 100,000+ engineers, and 10+ billion observations per month.

Tradeoffs are usage and governance limits. Hobby is capped at 50k units/month, 30 days of data access, and 2 users, while SSO, RBAC, private support, and scheduled exports require higher paid options.

Pricing & Value

• Hobby: $0/month, with 50k units/month, 30 days of data access, 2 users, limited platform features, and community support
• Core: $29/month, with 100k units/month, 90 days of data access, unlimited users, in-app support, and $8 per extra 100k units
• Pro: $199/month, with 100k units/month, 3 years of data access, high rate limits, retention controls, annotation queues, SOC2 and ISO27001 reports, and BAA availability
• Teams Add-on: $300/month, adding enterprise SSO, SSO enforcement, fine-grained RBAC, and dedicated Slack or MS Teams support
• Enterprise: $2,499/month, with Pro plus Teams, audit logs, SCIM API, custom rate limits, uptime and support SLAs, and dedicated support

The free plan suits prototypes and POCs, while paid plans buy longer history, more users, higher limits, support, and security controls.