Langfuse

Langfuse is an open-source AI engineering platform for improving LLM and agent applications. It combines traces, sessions, cost and latency analytics, prompt versioning, datasets, experiments, evaluations, feedback, and alerts in one API-first workspace. Teams can use Langfuse Cloud or self-host the MIT-licensed core.

✨ Key Features

• LLM and agent tracing — Capture model calls, retrieval, tools, embeddings, API calls, and agent graphs as connected observations.
• Sessions and users — Group multi-turn conversations and inspect usage by user, feature, release, or environment.
• Cost and latency analytics — Track tokens, spend, volume, latency, and quality in dashboards and the Metrics API.
• Prompt management — Create, version, compose, label, deploy, cache, and measure text or chat prompts.
• Playground and experiments — Test prompt versions and models against datasets, then compare results.
• Evaluations and scores — Use LLM-as-a-judge, code evaluators, human annotations, user feedback, or custom pipelines.
• Datasets and annotation queues — Build test sets, create ground truth, and review samples with structured workflows.
• SDKs and integrations — Instrument apps through Python/JavaScript SDKs, OpenTelemetry, LiteLLM, public APIs, and frameworks.
• Dashboards and alerts — Slice metrics by model, prompt, user, session, or tag and notify through Slack, GitHub Actions, or webhooks.
• Cloud and self-hosting — Use managed Cloud regions or deploy the open-source core with Docker Compose, Kubernetes, or cloud infrastructure.

💳 Pricing

• Hobby — Free, no credit card; 50k units/month, 30 days of data access, 2 users, and community support.
• Core — $29/month; 100k units/month, 90 days of data access, unlimited users, and in-app support. Additional usage starts at $8 per 100k units.
• Pro — $199/month; 100k units/month, 3 years of data access, higher ingestion limits, retention management, unlimited annotation queues, and prioritized support. Additional usage starts at $8 per 100k units.
• Enterprise — $2,499/month; enterprise support, audit logs, SCIM, custom limits, SLAs, and a dedicated support engineer. Yearly commitments can use custom volume pricing, AWS Marketplace, or invoicing.
• Self-hosted Open Source — Free under MIT for the core platform and APIs with unlimited included usage; infrastructure and optional Enterprise capabilities are separate.

Usage is measured in ingested traces, observations, and scores. Confirm current pricing, retention, regions, limits, and optional enterprise features on the official pricing page.

⚖️ Alternatives Compared

🎯 Use Cases

• Debugging agents — Follow tool calls, retrieval, generations, and latency in one trace.
• Prompt iteration — Version prompts and compare quality, cost, and latency on a dataset.
• RAG monitoring — Inspect retrieval context and score relevance or answer quality.
• Cost control — Break down token usage by model, feature, user, or session and alert on thresholds.
• Regression testing — Run SDK or UI experiments in CI/CD after model or prompt changes.
• Human review — Collect feedback, annotate samples, and build ground truth.
• Self-hosted observability — Keep the open-source core in your infrastructure when data control matters.

Langfuse supports engineering observability and quality workflows; it does not guarantee that an LLM output is correct. Verify retention, masking, access controls, model pricing, and evaluator behavior.