Langfuse

What it does

Langfuse takes in traces from LLM calls, tool invocations, retrieval steps, and other non-LLM calls, captured through native Python and JavaScript SDKs, OpenTelemetry, or an LLM gateway such as LiteLLM. It produces hierarchical traces, session and user views, cost and latency metrics, dashboards, and alerts, along with prompt versions, datasets, experiment results, and evaluation scores.

Teams use Langfuse to move from production debugging to repeatable testing: inspect failed or slow traces, turn findings into datasets, run experiments on prompts or models, and score outputs with LLM-as-a-judge, code evaluators, user feedback, or human annotation queues. Tracing, prompt management, evaluation, and experiments share one data model, and the core product is MIT-licensed with a self-hosting path.

Langfuse suits developers and engineering teams building LLM or agent applications who want control over their data. Cloud access starts with a free Hobby plan that includes 50k units per month and requires no credit card. Self-hosting is free to run as software, but you pay for your own infrastructure. The Hobby plan includes 30 days of data access and 2 users; paid tiers add usage fees and higher rate limits.

What it can do

• Hierarchical tracing of LLM calls, tool invocations, and retrieval steps with filters by user, session, cost, latency, or metadata
• Prompt management with versioning, deployment labels, and rollback without code changes
• LLM-as-a-judge, heuristic, and human annotation evaluation on production traces or experiments
• Datasets and experiments that compare prompt or model variants side by side
• Dashboards and alerts for cost, latency, and quality via Slack, GitHub Actions, or webhooks
• Playground for testing prompts and comparing models

Works with

• LangChain
• Vercel AI SDK
• LiteLLM
• OpenAI
• Anthropic
• Amazon Bedrock
• Google Gemini
• Claude Code
• PostHog

Example workflows

Debug a slow production agent

Input: production traces from an agent. Steps: filter traces by high latency, open the timeline view, and inspect each LLM and tool call. Output: identified slow step with its cost and latency metrics.

Compare two prompt versions offline

Input: a dataset of test cases and two prompt versions. Steps: run an experiment in the UI or via SDK, then score outputs with an evaluator. Output: side-by-side scores for each version.

Before you choose

• Hobby plan includes only 30 days of data access and 2 users; longer retention and unlimited users require Core or higher.
• Usage above included units is billed per 100k units with graduated rates, and Enterprise is priced separately via sales.
• Self-hosting is free to run as open-source software, but you pay for your own infrastructure and operations.

Listed under

• Coding
• DevOps and observability
• AI agents
• Developer tools
• Evals
• LLM observability
• Open source
• Prompt management

Get started

No credit card required. Setup uses the SDK, CLI, or MCP server.

1. Create a free Hobby account at langfuse.com/cloud and create a project.
2. Install the Python or JavaScript SDK and instrument your application following the get-started guide.
3. Optionally install the Langfuse Agent Skill, CLI, or MCP server so a coding agent can manage prompts, traces, and evals.

Hobby is free with 50k units per month included and no credit card required. Core costs $29/month with 100k units included, then $8 per additional 100k units with volume tiers. Pro is $199/month, and Enterprise starts at $2,499/month. Team add-on is $300/month. Figures are as published on the pricing page.

Source: langfuse.com/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Langfuse: Open Source Agent Evals & Observability ↗langfuse.com
• Langfuse joins ClickHouse - Langfuse ↗langfuse.com
• Pricing - Langfuse ↗langfuse.com
• Open Source AI Engineering Platform - Langfuse ↗langfuse.com

Company website

langfuse.com: product, pricing and policy pages

Hacker News

Launch posts and discussion

YC directory mirror

Batch and company name

Human review

Checked by a person

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.