The software around the model
LangChain builds tools for developers combining a language model with data, executable tools and application logic. Its current documentation separates the LangChain agent harness, where developers compose models, prompts and middleware, from LangGraph, the lower-level orchestration framework. That places the company at the application layer: it supplies ways to organize model calls and actions across providers such as OpenAI, Anthropic and Google. [1]
The distinction matters when an application needs both predictable code paths and choices delegated to a model. LangGraph exposes workflow state, branching and loops; LangChain offers a higher-level starting point for an agent. The documentation also describes Deep Agents, built on LangChain, as a more packaged option with context compression and a virtual filesystem. These are different levels of developer control within the same tool family. [1][3]
Making failures inspectable
LangSmith became generally available on February 15, 2024, after a closed beta that began in July 2023. The product records application traces and supports debugging, testing, evaluation and monitoring. At launch, LangChain emphasized inspecting the intermediate steps of an application and comparing behavior across datasets, rather than judging it only by a finished answer. [2]
This gives LangChain a role beyond distributing an open-source framework. A team can assemble an agent, then use the company's commercial tooling to investigate how it reached a result and whether a change helped. Traces and evaluations make behavior inspectable; the vendor's product announcement and selected customer accounts do not establish that any particular deployed agent is reliable. [2][1]
From integrations to longer-running work
For the Codestral release on May 29, 2024, Mistral named LangChain and LlamaIndex as community partners whose integrations were available that day. This is a concrete distribution role: developers could use the new code model through an existing application framework. Mistral remained the model developer. [4]
On June 27, 2024, LangChain released stable LangGraph v0.1 and introduced LangGraph Cloud in closed beta. The announcement described persistent state, recovery and human intervention for longer-running agents. It marks the company's move from connecting individual model calls toward managing the execution of an application over time; closed beta described the hosted service's access at that announcement, while the framework itself had a stable release. [3]
Sources
- LangChain overview
LangChain · retrieved Oct 23, 2026
- Announcing the General Availability of LangSmith and Our Series A Led By Sequoia Capital
LangChain · Feb 15, 2024
- Announcing LangGraph v0.1 & LangGraph Cloud: Running agents at scale, reliably
LangChain · Jun 27, 2024
- Codestral
Mistral AI · May 29, 2024