LangSmith
LLM observabilityProprietary SaaS, deepest inside the LangChain/LangGraph ecosystem. Configurable alerting, LangGraph Studio, and an Insights Agent that clusters production conversations into usage patterns and failure modes.
LangSmith vs Langfuse
LangSmith vs Langfuse.
And what reads the conversations.
Both are LLM observability tools for engineers, and both ingest OpenTelemetry. LangSmith is proprietary and built by the LangChain team; Langfuse has an open-source core you can self-host for free. Here's how they differ — and what sits on top of either.
01positioning
The LangSmith vs Langfuse decision usually comes down to ownership and ecosystem: LangSmith is a managed SaaS built around LangChain and LangGraph, while Langfuse is open source (MIT), self-hostable, and framework-agnostic. Both trace what the system did, and LangSmith now clusters conversations by intent and failure mode on top of that. Neither rolls the result into a single score for the agent, or proves a fix worked.
Langfuse
Open-source LLM observabilityOpen source and self-hostable, framework-agnostic, transparent unit pricing. Strong prompt, eval, and dataset workflows.
brizz
Agent analyticsWhichever tracer you pick, it records what the system did. Brizz is the analytics layer that reads what users wanted and whether the agent delivered — detecting intents, scoring each issue's impact, and proving the fix worked, for the whole team rather than just engineers. Brizz ingests the same conversations over OpenTelemetry, so it works with either.
02the decision
Head to head, on the facts.
| Criterion | LangSmith | Langfuse |
|---|---|---|
| License | Proprietary. The client SDKs are MIT. | MIT core. Enterprise modules (SCIM, audit logs, data retention) need a commercial license. |
| Self-hosting | Enterprise add-on, on Kubernetes with Helm. A hybrid option keeps the data plane in your cloud. | Free to self-host with Docker Compose, Kubernetes or Terraform. Enterprise features are paid. |
| Cloud regions | US, EU and APAC. | US, EU and Japan, plus a separate HIPAA region. |
| Pricing model | Per seat plus traces. Free for 1 seat and 5k traces a month; Plus is $39 per seat a month with 10k traces. | Per unit, no seat fees. Free for 50k units a month; Core is $29 a month with 100k units and unlimited users. |
| OpenTelemetry | Native OTLP endpoint. Maps GenAI, OpenInference and Traceloop attributes. | Native OTLP endpoint over HTTP (no gRPC yet). The SDKs are built on OpenTelemetry. |
| Frameworks and SDKs | Works beyond LangChain, deepest inside LangChain and LangGraph. SDKs for Python, TypeScript, Java and Kotlin. | Framework-agnostic. SDKs for Python and JS/TS, other languages via OpenTelemetry. |
| Evaluations | Datasets and experiments, online LLM-as-a-judge including multi-turn, pairwise comparison, annotation queues. | Datasets and experiments (including in CI), online LLM-as-a-judge, code evaluators, annotation queues. |
| Prompt management | Versioned prompts with commits and tags, plus a Playground. | Versioned prompts with labels, a playground, prompt experiments and a GitHub integration. |
| Alerting | On run count, cost, errors, feedback and latency, to Slack, PagerDuty, Dynatrace or webhooks. | On latency, cost, count and scores, to Slack, webhooks or GitHub Actions. Capped per plan on cloud. |
| Clustering production traces | Insights reports cluster traces into usage patterns and failure modes, on demand or on a schedule (Plus and up). | No automatic clustering. Set up categorical LLM-as-a-judge scores, or ask the in-app Assistant. |
Last reviewed September 2026 · from each vendor's docs and pricing page
03capabilities
Three tools, side by side.
Capability
LangSmith
Langfuse
brizzSingle Agent Health Score
×
×
✓
Closed loop, verify the fix worked
×
×
✓
Automatic issue detection with impact scoring
●
×
✓
Built for PM, exec, and builder
●
●
✓
Framework-agnostic beyond LangChain
●
✓
✓
Automatic semantic intent detection
✓
×
✓
Native alerting
✓
✓
✓
LLM tracing and spans
✓
✓
●
Evals and LLM-as-judge
✓
✓
●
Open source and self-hostable
×
✓
×
Full Partial Not offered
Last reviewed September 2026 · from each vendor's public docs
04where brizz fits
LangSmith or Langfuse handles tracing and evals. Brizz turns those same conversations into product decisions. Pick a tracer — and add Brizz on top.
New to agent analytics? Start here→LangSmith vs Langfuse, answered
Choose Langfuse for open-source self-hosting, data sovereignty, and a framework-agnostic approach. Choose LangSmith if you're all-in on LangChain/LangGraph, or if you want conversation clustering built in — its Insights Agent groups production traces into usage patterns and failure modes without you defining a taxonomy, where Langfuse expects you to configure that yourself as a categorical LLM-as-a-judge evaluator.
Langfuse is open source (MIT) and self-hostable as a first-class option. LangSmith is proprietary SaaS and requires an enterprise license to self-host.
Brizz adds the analytics layer: automatic intent and journey detection, issue impact scoring, a single Agent Health Score, and closed-loop verification — read from the same conversations your tracer already captures.
No. They trace the runtime for engineers; Brizz reads meaning and outcomes for the whole team. Keep your tracer and add Brizz on top.
See it on your own agents.
Brizz turns every agent conversation into intel your whole team can act on.