Langfuse v4 is live

Langfuse v4 is live on Langfuse Cloud. It makes tables, dashboards, APIs, and evaluations dramatically faster at scale. Dashboard load times for large projects improve by at least 10x.

At its core is an observations-first data model built for agentic systems. Query any LLM call, tool execution, or agent step directly, without costly joins or read-time deduplication.

What’s new

Every operation is now a first-class queryable object. Find slow LLM calls, expensive models, failed tool calls, or individual agent steps directly across any number of traces.

Under the hood, Langfuse stores one immutable observations table. A trace is all rows that share a trace_id, and trace-level attributes are copied onto every row:

Observations table

Observation-level data				Trace-level data · on every row
id	type	name	latency	trace_id	trace_name	user_id	session_id
obs-1	span	handle-chat	3.1s	abc123	chat-request	u-42	s-7
obs-2	span	retrieval	0.4s	abc123	chat-request	u-42	s-7
obs-3	generation	llm-call	2.6s	abc123	chat-request	u-42	s-7
obs-4	span	build-report	1.2s	def789	export-report	u-17	s-9
obs-5	generation	summarize	0.9s	def789	export-report	u-17	s-9

Trace-level input/output is deprecated across the product. Tables, LLM-as-a-judge evaluators, and exports now read input and output from the relevant observation, usually the root span or LLM call.

Selecting one observation instead of assembling data across an entire trace is more precise and a major reason v4 loads faster.

Observations-first lets you query and evaluate the relevant sub-agent, tool call, or LLM call directly. Immutable observations also avoid repeated trace-output updates and costly read-time deduplication.

Learn more in the data model docs and the observations guide.

Timeline

On Langfuse Cloud, review your project's remaining migration actions in the app.

Upgrade

Follow the step-by-step upgrade guide. It covers Langfuse Cloud, direct API and OpenTelemetry users, and self-hosted deployments.

Check the Versions & Compatibility matrix to see which SDK and server versions work together.

Questions

Ask rollout and migration questions in the dedicated GitHub Discussion; we use that thread for updates and answers.

---

---