LangSmith
LangSmith (LangChain) provides tracing, evaluation, deployment sandboxes, and an language model gateway for agent stacks with SDKs in Python, TypeScript, Go and Java. It supports hosted and self-hosted trace storage and meters usage in LangChain Compute Units (LCUs) and Storage Units (LSUs); outputs are traces, evaluations, and deployed agents.
Developer: $0 / seat per month with up to 5k base traces/month included; Plus: $39 / seat per month with 10k base traces/month included; Enterprise: custom pricing and self-host options. Many features are metered in LCUs and LSUs with pay-as-you-go billing.Is LangSmith right for you?
Good for
- Teams building and debugging production agents who need end-to-end traces, multi-language SDKs, and a choice of hosted or self-hosted deployments.
- Multi-language SDKs let you capture traces from Python, TypeScript, Go, or Java agent stacks.
- You can self-host trace storage (SmithDB) inside your VPC so sensitive traces do not leave your infrastructure.
Keep in mind
- Some evaluation features are initially limited to public beta for certain plans and regions.
- Traces have different retention tiers (base vs extended) that affect storage and cost.
- Buyers seeking simple fixed per-seat pricing without usage metering should prepare for metered billing based on traces, LCUs and LSUs and confirm cost expectations beforehand.
Pricing and plan limits
The pricing pages list seat prices and included base traces, LCU and LSU unit prices, and metering rules. Specific overage rates for traces and detailed Enterprise invoices require contacting sales or consulting the docs.
More about capabilities and limits
Documented strengths
Multi-language SDKs let you capture traces from Python, TypeScript, Go, or Java agent stacks.
You can self-host trace storage (SmithDB) inside your VPC so sensitive traces do not leave your infrastructure.
Billing and compute are explicit: compute work and storage are metered as LCUs and LSUs with stated unit prices.
Limitations to consider
Some evaluation features are initially limited to public beta for certain plans and regions.
Traces have different retention tiers (base vs extended) that affect storage and cost.
Choosing a plan
Estimate your trace volume and storage needs, then map expected usage to LangSmith’s LCU/LSU metering to forecast monthly costs and choose hosted vs self-hosted trace storage.
Based on the linked product documentation.
Available on
Web.
Official sources
Alternatives to LangSmith
Choose around the work you need to do.
AgentOps
Use AgentOps to record, inspect, and replay agent runs so developers can trace tool and model calls, debug failed executions, and track tokens and spending during multi-step workflows.
Langfuse
Choose Langfuse to trace model calls, inspect latency and cost, and run prompt/application evaluations with an open-source, self-hostable observability platform for language-model applications.
Braintrust
Pick Braintrust to get end-to-end tracing of nested agent traces plus automated scoring and dashboards for monitoring latency, cost, and output quality in production language model agents.