PrefactorvsLangfusevsLangSmith

Langfuse vs LangSmith: the honest version

LangSmith if you are all in on LangChain, Langfuse if you want open-source self-hosting. Neither alone answers whether the agent did its job; that is where Prefactor sits on top of, or instead of, both.[1][3]

support-agent v4 · one run, two layersexample
Illustrative run, showing what each layer tells you
Langfuse / LangSmithsee
fetch_customer212ms · 1.2k tok
apply_refund1.4s · 3.1k tok
send_reply340ms · 0.8k tok
trace recorded, no verdict
Prefactoradds
Did its job✓ yes
Quality84 / 100
Cost$0.42 · in budget
Drift vs baselinenone
a run that breaches its schema is held for review
§01 / THE SHORT ANSWERtl;dr: which, and when
TL;DR

Pick LangSmith if you build on LangChain or LangGraph and want first-party agent detail; pick Langfuse for open-source tracing you can self-host, on any framework. Both stop short of a verdict on the whole agent session, which is the layer Prefactor adds on top of either.

The short answer

The quick decision

Decision factorLangfuseLangSmith
Ecosystem couplingFramework-agnostic SDKs, no preferred frameworkWorks anywhere, strongest with LangChain and LangGraph
Open source / self-hostMIT-licensed core, self-hostableProprietary, self-hosted on enterprise plans only
Tracing depthStrong traces, sessions, multi-modal loggingDeepest LangGraph agent detail, step-level debugging
EvalsLLM-as-judge evaluators, annotation queues, scores on tracesDatasets, experiments, LLM and heuristic evaluators, online evals
Cost attributionToken and cost per trace and user, per-agent is yours to buildToken and cost per run, per-agent is yours to build
Pricing modelOpen source free, cloud platform fee plus usage per unitPer seat plus per trace, retention-based tiers
Direction riskAcquired by ClickHouse in January 2026, roadmap questions in flightTied to LangChain's commercial trajectory and its ecosystem bets
§02 / IN DEPTHdetail: the reasoning
In depth

Langfuse in depth

Langfuse is the strongest open-source option in LLM observability. The core platform, tracing, sessions, prompt management with versioning, cost analytics, evaluators, and annotation queues, is MIT-licensed and self-hostable, which makes it the default for teams with data-residency requirements or an aversion to per-trace metering. The SDKs are genuinely framework-neutral: LangChain, LlamaIndex, the OpenAI SDK, or raw HTTP all instrument cleanly, and OpenTelemetry compatibility keeps exit costs low.

Limits, fairly stated: the evaluation features are trace-centric, good for scoring individual generations, lighter for multi-step task verdicts and eval-gated deploys. Dashboarding is functional rather than deep, and anything organisation-specific, such as per-customer cost rollups or business KPI joins, means exporting the data and building it yourself. Since January 2026, Langfuse is part of ClickHouse; the team has said the open-source product continues, and the open questions are roadmap coupling to a ClickHouse-native architecture and how cloud and residency options evolve. That is a reason to keep your quality layer portable, not a reason to leave.

LangSmith in depth

LangSmith is the commercial observability and evaluation platform from the LangChain team. Its defining advantage is first-party depth: if your agents are built on LangChain or LangGraph, instrumentation is automatic and the trace view understands your agent's structure, graph nodes, tool calls, and retries, rather than showing generic spans. The evaluation tooling (datasets, experiments, online evaluators) and the prompt playground are polished, and the hosted product removes operational burden.

The reality check on LangChain coupling: LangSmith does work without LangChain, through OpenTelemetry ingestion and standalone SDKs, but most of what makes it better than alternatives assumes LangChain primitives. Adopting it from outside that ecosystem buys a good but not exceptional tracer at proprietary pricing. It is not open source, self-hosting is gated to enterprise contracts, and per-seat-plus-per-trace pricing needs modelling before you commit high-volume agent traffic to it.

What they cost at 1M traces a month

Published list prices as of June 2026, and both vendors revise pricing, so treat these as order-of-magnitude and verify on their pricing pages. LangSmith meters per trace, at a list price around $0.50 per 1,000 base traces on 14-day retention and several times that for extended retention, plus per-seat fees on paid plans. At 1M base traces a month that is roughly $500 a month in trace charges before seats and retention upgrades.

Langfuse Cloud charges a platform fee per tier plus tiered usage per ingested unit, and an agent session typically emits several units. Comparable volumes usually land in the low hundreds of dollars a month on the published tiers. Self-hosted Langfuse has no usage bill at all: you pay in infrastructure, since it runs on ClickHouse and Postgres, and in the engineering time to operate it, which is rarely zero. At small volume the difference is noise; at serious agent volume, per-trace metering compounds fastest and self-hosting is cheapest in cash and most expensive in attention.

§03 / THE QUALITY GAPour take: where it stops
Our take

Where both stop: the quality gap

We sell the layer this section describes. Read it with that in mind.

Both products answer what your LLM application did, and they answer it well. Neither is built to answer whether each agent did its job, at acceptable quality and cost, with proof you can show.

01
Task-completion verdicts

A verdict for the whole multi-step agent session rather than per-generation scores, alongside evals that run on the traces you already collect.

02
Feedback tied to the agent

Customer and business feedback captured where it happens and tied to the agent that caused it.

03
Cost per agent and version

Cost attributed per agent, per version, and per customer rather than per trace, the basis of per-agent analytics.

04
An audit-grade record

Quality over time in a record you can hand to a customer or a regulator. Prefactor ingests traces from Langfuse, LangSmith, or any OpenTelemetry source, so whichever tracer you choose, you keep it.

See it on your own agents

A working session on a fleet like yours: watch a run evaluated, catch a drift, walk the record.

§04 / WHICH TO PICKdecide: by your stack
Which to pick

Which one fits

Pick Langfuse if

  • Open source or self-hosting is a requirement.
  • You use multiple frameworks, or none.
  • You want prompt management without vendor lock-in.
  • Data residency drives your architecture.

Pick LangSmith if

  • You are committed to LangChain and LangGraph.
  • You want first-party agent step detail.
  • Hosted-only is acceptable, or enterprise budget exists.
  • Your eval workflow lives in datasets and experiments.
§05 / HOW WE REVIEWEDsources: checked June 13, 2026
Methodology

How we reviewed this comparison

Reviewed against public product and documentation pages on June 13, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.

Sources reviewed

  1. Langfuse documentation
  2. Langfuse pricing
  3. LangSmith documentation
  4. LangSmith pricing
  5. Langfuse is joining ClickHouse (announcement)Acquisition announced January 2026; referenced in the post-acquisition risk discussion.
Prefactor context

Methodology

  • Reviewed public product, documentation, and launch material visible at the time of writing.
  • Mapped each page to the primary buyer, control layer, and runtime capabilities each vendor describes publicly.
  • Prefer direct product and documentation pages over analyst summaries or reseller material.
§06 / QUESTIONSfaq: the common ones
Questions
Langfuse vs LangSmith: which is better for agents?
For agent tracing both are capable: LangSmith has the deepest read on LangGraph agents because the same team builds it, while Langfuse traces agents from any framework and lets you self-host. Neither is better for agents universally; the deciding factor is your stack. For agent quality, meaning completion verdicts, business feedback, and per-agent cost, both stop short, and that is the layer a quality platform adds on top.
Is LangSmith only for LangChain?
No. LangSmith supports OpenTelemetry-based ingestion and SDKs that work without LangChain, but its value concentrates in the LangChain and LangGraph ecosystem, where automatic instrumentation, agent step views, and prompt tooling are at their best. Teams on other frameworks can use it, but they give up much of what makes it compelling.
Is Langfuse open source? Is LangSmith?
Langfuse is open source: the core platform is MIT-licensed and self-hostable, with some enterprise features under a commercial licence, and the project continues after the January 2026 ClickHouse acquisition. LangSmith is proprietary: a hosted commercial product, with self-hosted deployment on enterprise plans and no open-source edition.
Do I still need evals if I have tracing?
Yes. Tracing records what the agent did; evals judge whether it did its job. A trace of a hallucinated answer looks identical to a trace of a correct one: same spans, same latency, same token counts. Both tools ship eval features attached to traces, which is a good start, and a production agent additionally needs golden-dataset regression on every change and scored sampling of live traffic, tracked per agent version over time.
Can I use one of them and Prefactor together?
Yes. Keep Langfuse or LangSmith as your tracer and let Prefactor read those traces to add completion verdicts, evals, feedback, and per-agent cost. Prefactor ingests any OpenTelemetry source, so you keep the tracer you chose.
Reviewed against public sources on June 13, 2026Suggest a correction

Get the quality layer that works with either

Prefactor adds completion verdicts, evals, feedback, and per-agent cost on top of the traces you already collect, from Langfuse, LangSmith, or any OpenTelemetry source. Book a demo to see it on a fleet like yours.

Agent Performance Platform
Unified performance platform for agents, authentication, and risk management
All Systems Operational
3Global Agents
7Instances
5Services
12%Human Intervene
4High Risk
$2,360Monthly Spend
Mission ControlLive agent health with 7-day activity heartbeat
Claims Proc...68
$330/moRed
Claims Proc...65
$160/moRed
Claims Proc...82
$170/moAmber
ChatGPT74
$150/moAmber

See how every agent performs, and make it better

Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.