LangSmith+17 Mixed

GILD Review

GILD's call and the buying facts behind it.

From GILD research

Should we adopt LangSmith?

Is LangSmith worth buying if your agents do not run on LangGraph?

30-Second TakePick LangSmith if your agents already serve users and you want traces, experiments and hosting from one vendor.
Best fitTeams shipping agents to production, especially on LangGraph, that want tracing and evals from the same vendor that hosts their agents.GILD Review
Consider another workflowIf you need an open-source platform you can self-host without an Enterprise contract, evaluate Langfuse; if you also monitor classic ML models, evaluate Arize AX.GILD Review

Drawn from 125 sources and 9 operator insights

GILD analysis

LangSmith Review

Pick LangSmith for tracing, offline and online evaluators, and agent hosting from LangChain, with each change scored against datasets built from your own production failures. LangChain and LangGraph integrate directly; other code connects through SDKs or OpenTelemetry. Base retention lasts 14 days, billing is per trace, and self-hosting needs an Enterprise contract.

AI Observability
AI Costs

The decision

Best fit: Teams shipping agents to production, especially on LangGraph, that want tracing and evals from the same vendor that hosts their agents.
Consider another workflow: If you need an open-source platform you can self-host without an Enterprise contract, evaluate Langfuse; if you also monitor classic ML models, evaluate Arize AX.
Adoption condition: A default retention tier per project, masking for sensitive inputs, and a monthly usage limit, all set before production traffic is traced.

Basis: LangSmith documentation, plans, and pricing, plus 110 public accounts, among them Parlance Labs' 100-trace test of LangSmith Engine, Hamel Husain's side-by-side evaluation panel and 33 GitHub issues and pull requests. In a pilot, compare app memory and request latency with tracing on and off.

Buying & deployment factsWhat a team building on it should verify before production traffic grows.

How you buy it

US$39

Per seat per month for Plus, with 10,000 base traces. Developer is free for one seat.

Source 1, langchain.com, Plans and pricing: Plans and pricing

Details

Developer costs US$0 for 1 seat and 5,000 monthly base traces. Plus costs US$39 per seat per month, with 10,000 monthly base traces and 1 free Small Serverless deployment. Enterprise uses custom annual invoices. Self-serve seats bill monthly, with trace usage billed monthly in arrears.

Cost drivers

US$5 per 1,000

Base traces past the allowance, kept 14 days. Evaluators upgrade traces by default.

Source 1, langchain.com, Plans and pricing: Plans and pricing Source 2, LangChain, Usage and billing: Usage and billing

Details

Beyond the included volume, a base trace with 14-day retention costs 0.005 LSU, or US$5 per 1,000. Extended retention lasts 180 days and adds 0.0025 LSU per trace. 1 LSU equals US$1. Online evaluators and automation rules upgrade traces by default. Deployment, Engine, Fleet, and Sandboxes also bill in LSUs.

Where it runs

US, EU, APAC

LangChain Cloud regions on GCP, plus AWS US. Enterprise can self-host on Kubernetes.

Source 3, LangChain, Cloud regions: Cloud regions Source 4, LangChain, Self-hosted LangSmith: Self-hosted LangSmith Source 5, LangChain, LangSmith Deployment: LangSmith Deployment Source 6, LangChain, Trace with OpenTelemetry: Trace with OpenTelemetry

Details

LangChain Cloud runs in GCP US, EU, and APAC regions and an AWS US region. Enterprise adds self-hosting on Kubernetes. Agent workloads also support hybrid and BYOC. Tracing connects through Python and TypeScript SDKs, LangChain integrations, or OpenTelemetry.

What to verify

14 days

Base trace data is deleted then, but some metadata is kept indefinitely.

Source 1, langchain.com, Plans and pricing: Plans and pricing Source 2, LangChain, Usage and billing: Usage and billing Source 7, LangChain, Prevent logging of sensitive data in traces: Prevent logging of sensitive data in traces

Details

LangChain says it does not train on traces. It deletes base trace data after 14 days but keeps some metadata indefinitely for analytics and billing. Dataset examples never expire. The SDK can hide or mask inputs and outputs before export.