LangChain

What the company is

LangChain builds two open-source frameworks, LangChain and LangGraph, and a commercial platform called LangSmith ^[1]. The frameworks are software libraries that developers use to assemble applications on top of large language models; LangSmith is the paid product that surrounds them.

The company describes the work its platform serves as agent engineering: the iterative refinement of non-deterministic LLM systems into reliable experiences ^[1]. The phrase points at a practical difficulty. A conventional program produces the same output for the same input, while a system built on a language model may not, so getting such a system to behave consistently takes repeated testing, measurement and adjustment rather than a single act of writing code.

Products

The company's site groups its offerings as Deep Agents, LangGraph, LangChain and Fleet, with LangSmith Engine for insight ^[2].

LangSmith itself spans observability, evaluation and deployment ^[1]. Observability means recording what an application actually did when it ran, so that developers can look back at the individual steps; evaluation means measuring whether those runs were good. The former LangGraph Platform, the part of the product concerned with putting an application into service, was renamed Deployment ^[1].

Deployment runs on infrastructure built for long-running tasks, with more than 30 Agent Server endpoints and isolated sandboxes for code generated by agents ^[2]. Long-running tasks and agent-written code both create their own requirements: work that continues over an extended period needs somewhere to keep its state, and code produced by a model at run time needs to be separated from everything else.

Funding and scale

On 20 October 2025 the company announced it had raised $125M at a $1.25B valuation ^[1]. IVP led the round, alongside existing investors Sequoia, Benchmark and Amplify, and new investors CapitalG and Sapphire Ventures ^[1].

The company also gave figures for its reach at that point. It said its frameworks then had a combined 90 million monthly downloads, that 35 percent of the Fortune 500 used its services, and that monthly trace volume on LangSmith had grown 12 times year over year ^[1]. Trace volume measures how much recorded activity passes through the platform, so the figure describes usage of the observability side of the product rather than downloads of the open-source code.

Named customers include Replit, Clay, Harvey, Rippling, Cloudflare, Workday and Cisco ^[1].

Where it sits

LangChain occupies an unusual position for a company of its size, because its most widely used pieces are open-source frameworks that anyone can download, while its revenue comes from a commercial platform built around them. The two are connected: the more applications are built with the frameworks, the more reason there is to record and evaluate what those applications do, which is the job LangSmith is sold to do.

The emphasis on observation and evaluation reflects what makes software built on language models different from earlier kinds of software. When output is not fully predictable, the questions a team needs to answer shift from whether the code is correct in principle to what the system actually produced across many runs, and whether that was good enough. A platform that records runs and scores them is an attempt to make those questions answerable in an ordinary development workflow.

That is also why the company frames its subject as a discipline rather than a single tool. Agent engineering, in its usage, is the ongoing work of turning a system that behaves unpredictably into one that behaves reliably, and the products are organised around the stages of that work: building with the frameworks, watching and evaluating with LangSmith, and deploying on infrastructure designed for agents that run for a long time and write their own code ^[1][2].