LangChain

LangChain is an open agent platform for developers, with a commercial product called LangSmith on top and a no-code agent builder for people who don't write code. Its home page says it is "the open agent platform to own your intelligence", and the company's about page puts the mission as enabling every company to own its intelligence. This is mainly an engineering tool. Its no-code side is aimed at business colleagues who want an agent for their own daily routines, not at someone dropping a chat bubble onto a website.

What you build and how

The company splits its work into open source frameworks and a commercial platform. The about page says it develops open source projects to keep pace with new patterns for building agents, then folds what it learns into LangSmith, "our commercial agent engineering platform". The open source side has three projects named on the site. Deep Agents builds long-running agents for complex tasks, LangGraph builds agents with low-level control, and LangChain is for quick-start agents with any model provider. These are code libraries.

LangSmith has three groups of products. The first is the platform: an agent improvement engine that improves agents autonomously, observability to "see exactly what your agents are doing", and evaluation to score and improve agent performance. The second is agent infrastructure: deployment to ship and scale agents in production, sandboxes to run agent-generated code safely, and an LLM gateway to control agent model calls. The third is no-code agents, under the name Fleet, described as agents for the whole company. The pricing page also mentions serverless deployment, and traces, which are the recorded runs that observability reads.

The no-code path is where this entry matters for chatbot and agent builders. In its blog post, the company announces LangSmith Agent Builder in private preview, a "no code agent-building experience, complete with memory and guided prompt creation". It says it has spent the past three years building agents alongside millions of developers, and that engineering teams keep telling it how much their colleagues want to build their own.

The post is explicit about what the tool is not. LangSmith Agent Builder "is an agent builder, not a visual workflow builder", and the company gives two reasons: a visual workflow builder is not a low barrier to entry, and complex tasks quickly become too complicated to manage in one. Instead, an agent is guided by a prompt and can delegate more to the model than a predetermined path would. Guided prompt creation helps you write effective prompts without prompt engineering expertise. The agent has built-in memory, covering not only its prompt but also the tools it and any subagents can use, and it can update that memory at any point. If you correct the agent, it will remember the correction so you don't repeat it.

The company says Agent Builder is great for internal productivity use cases like email, chat and Salesforce assistants. One example it gives is an agent that sends you a summary of your schedule with meeting prep. That reads as a personal or team assistant, not a public support bot.

On the pricing page the no-code product now sits under the Fleet name, described as a way to "Create agents for daily tasks using everyday language, not code". The page lists access to prebuilt templates, sharing and cloning agent templates, unlimited agents, adding tools through a remote MCP server, and triggering an agent via API. Fleet models come in three performance tiers, Fast, Pro and Max, and the FAQ says these are tiers rather than fixed model names: LangChain picks and maintains the underlying models, so you do not need your own model-provider API key. Third-party tools usually need you to sign in with your own account, and any charges for them are billed by that provider. Every Fleet run is traced in LangSmith automatically.

For voice, the docs describe voice agents as agents that engage in spoken conversation by combining speech recognition, language processing, generative AI and text-to-speech. A guide builds a sandwich-shop ordering agent using what it calls the sandwich architecture, with AssemblyAI for speech to text and Cartesia for text to speech, and WebSockets for real-time communication between the browser and server. The docs say this architecture can achieve sub-700ms latency to support natural conversation, and that adapters can be built for most providers. A reference application is available in the voice-sandwich-demo repository.

What the company's site states

The pricing page lists three plans. Developer is $0 per seat, Plus is $39 per seat, and Enterprise is custom pricing. Usage is billed on top as pay-as-you-go: LCU, LangChain Compute Units for work done and compute, at $1.50 per LCU, and LSU, LangChain Storage Units for traces and storage, at $1.00 per LSU. A usage estimator on the page shows an example monthly total. A seat is a distinct user in your organization, and invited users count toward the total.

Billing rules are written out. For the Developer or Plus plan, seats are billed monthly on the 1st, or pro-rated if added mid-month, with no credit for removed seats, and traces are billed monthly in arrears for your usage. Enterprise plans are invoiced annually upfront. The company says it offers a Startup Plan with discounted rates and generous free trace allotments for early-stage companies. It also says LangSmith does not use your data to train models, and that your traces, prompts and outputs remain private to your organization. The pricing FAQ mentions serverless deployment included and says the Enterprise plan is for teams needing more advanced administration, security, support or deployment options.

The home page claims 350M+ monthly open source downloads, 7K+ active LangSmith customers and 5 of the Fortune 10 as LangSmith customers. Fleet usage is metered in LCUs, with model costs included. The Developer plan includes 5 LCUs of Fleet usage per organization each month, after which Fleet pauses new runs until the allowance resets or the organization upgrades; Plus includes 25 LCUs a month and bills additional usage. The FAQ says Fleet can be self-hosted through the sales team.

The plan cards add more detail. Developer is for solo users, with up to 5k base traces a month, community support and one seat. Plus adds up to 10k base traces a month, access to Deployment and Engine, unlimited seats and one free small serverless deployment. Enterprise lists self-hosted and hybrid deployment options, custom SSO, ABAC and RBAC, a support SLA and custom seats and workspaces. Base traces are kept for 14 days and extended traces for 400 days. Cloud data sits in LangChain's cloud in the US or EU, according to the hosting table, while the self-hosted option runs in your own VPC. For VC-backed startups, the page offers discounted rates and up to $10,000 in credits.

Its footer lists a Trust Center, docs, a changelog, support and LangChain Academy. The community Slack is described as a place for open discussion, events, job opportunities and showing off agents, and it says it is not for product support, which goes to the LangChain Forum.

Who it suits and who it does not

LangChain suits developers who want code-level control and open source building blocks, and companies that want an engineering platform for observing, evaluating and deploying agents they build. It also suits an operations or support team inside such a company that wants Agent Builder to make its own email, chat or Salesforce assistant without waiting on engineering.

There are limits. Agent Builder was announced in private preview in the company's blog post, so availability and features may differ from what you can use today. It is built for agents that work for your staff rather than for a branded chat widget on a public website. The pricing mixes per-seat fees with LCU and LSU usage, which takes more effort to forecast than a flat plan. And the lack of a visual flow editor is deliberate, so teams who want to drag a conversation tree together will prefer a conventional chatbot builder.