EditorialWe wrote this page from the company's own website. The company did not ask for it and did not pay for it. Every claim links to the page it came from.
LangChain offers LangSmith, an agent engineering platform for building, testing, deploying and improving AI agents.[1][3]
EditorialUpdated 19 Nov 2026Written by Beacon from LangChain's website; passed our listing checks (sources, claims, company domain)Not claimed by the companyMade by LangChain
What it does
LangChain describes LangSmith as an agent engineering platform that covers every step of the agent development lifecycle.[1]
LangSmith records traces. A trace is one run of your application and can include many steps, such as LLM calls.[2]
It supports evaluation of agents. Its Tuned Evaluators feature is in public beta.[2]
It can deploy agents on managed Agent Servers. The company mentions 1-click deployments, built-in APIs and autoscaling.[2][3]
A product called Fleet runs agents with built-in tools. It can also connect third-party tools through a remote MCP server.[2]
Who it is for
The company says the Developer plan suits personal projects. It says the Plus plan suits teams that want to self-serve with moderate usage and collaboration needs.[2]
The Enterprise plan is aimed at teams that need more advanced administration, security, support or deployment options.[2]
A Startup Plan with discounted rates is offered to early-stage companies building agentic applications.[2]
In a LangChain blog post, Rippling, a workforce management platform, says its Rippling AI runs on LangChain Deep Agents and LangSmith. The post says Rippling shipped it in roughly 6 months.[4]
How it works
Base traces are kept for 14 days and extended traces for 180 days. You can upgrade base traces to extended traces for an extra fee.[2]
Usage across LangSmith products is measured in LangChain Standard Units (LSUs). 1 LSU equals $1.[2]
Serverless Agent Servers scale to zero when idle. Dedicated Agent Servers are always on, and the company recommends them for customer-facing agents.[2]
Every Fleet run is traced in LangSmith automatically. These traces count toward usage-based billing.[2]
In the Rippling example, a supervisor agent coordinates 5 to 7 specialised subagents. LangSmith handles tracing, evaluations and production monitoring.[4]
Pricing
Developer plan: 1 free seat and up to 5k base traces a month, then pay-as-you-go. It comes with community support.[2]
Plus plan: you can buy unlimited seats, with up to 10k base traces a month included, then pay-as-you-go. It includes 1 free small serverless deployment.[2]
Enterprise plan: pricing is through the sales team, and it is invoiced annually upfront. It offers self-hosted and hybrid deployment, and custom SSO, ABAC and RBAC.[2]
On Developer and Plus, seats are billed monthly on the 1st or pro-rated if added mid-month. Traces are billed monthly in arrears.[2]
Fleet includes a monthly free allowance of 7 LSUs per organisation on Developer and 37 LSUs on Plus. On Developer, Fleet pauses new runs once the allowance is used.[2]
Before you buy
Costs depend on usage. The company says its usage estimate is only an estimate, and actual LSU use varies with the work your agents do.[2]
There is no credit for seats you remove mid-month.[2]
Tuned Evaluators are in public beta for Plus and Cloud Enterprise plans in the US only.[2]
Fees for third-party tools connected to Fleet are billed by those providers.[2]
The company says LangSmith does not use your data to train models.[2]
Alternatives
Other developer tools tools we have pages for. We do not rank them.