LangSmith

LangSmith is an observability and evaluation platform from LangChain, built for teams developing and running applications powered by large language models. It's designed for engineers building LLM-based agents and applications — people who need real visibility into model behavior, repeatable evaluation workflows, and the infrastructure to actually deploy and manage agents once they leave the prototype stage. This is not a tool aimed at business users or non-technical stakeholders.

Pricing

LangSmith offers three tiers: a free single-seat Developer plan, Plus at $39/seat/month, and a Custom-priced Enterprise tier. Beyond each tier's base allowance, usage is metered through LangChain Compute Units ($1.5/LCU) and Storage Units ($1/LSU), so actual monthly cost scales with tracing volume and storage on top of the named plan.

SOC 2 Status

LangSmith's SOC 2 certification status isn't publicly disclosed in available information.

API Access

LangSmith exposes API access, so tracing instrumentation, dataset management, evaluation orchestration, and prompt management can all be wired directly into existing development and CI/CD pipelines — teams aren't limited to working through the web UI.

Core Features

LangSmith's features cover the entire lifecycle of an LLM application, from early debugging through production operation:

• Tracing and monitoring — captures execution traces of LLM calls and chains for debugging and performance inspection.
• Online and offline evaluations — supports evaluation workflows both against live traffic and against static datasets.
• Dataset collection and annotation queue — provides tooling to gather examples and route them through human annotation workflows.
• Prompt Hub and Playground — centralizes prompt versioning and offers an interactive environment for iterating on prompts.
• LangSmith Deployment (Serverless and Dedicated) — offers two deployment modes for running agents in production infrastructure.
• LangSmith Fleet — enables agent creation through natural language instructions.
• LangSmith Engine — performs automated root cause diagnosis and generates fixes when issues are detected.
• LangSmith Sandboxes — provides isolated environments for executing code safely, relevant to agentic workflows that generate and run code.
• LangSmith LLM Gateway — sits in front of model calls to provide cost controls, rate limiting, and PII redaction, addressing governance concerns for teams routing traffic through multiple models or providers.
• Bulk data export — allows extraction of traces, datasets, and other stored data out of the platform.

Summary

Between tracing, evaluation, deployment, and gateway-level governance, LangSmith reads as an end-to-end platform built for teams that are past the prototyping phase and need real operational tooling around their LLM applications — especially agentic systems that need failure diagnosis, contained code execution, and cost oversight across model calls. Starting at $0 keeps the barrier to entry low, but the hybrid pricing underneath means actual monthly spend will climb with LCU and LSU consumption as tracing volume, storage, and evaluation frequency grow — that's worth budgeting for up front rather than discovering later. Since SOC 2 status isn't publicly disclosed, teams with strict compliance requirements around data handling should get certification details directly from LangChain before putting sensitive workloads through the platform. And because API access is available, LangSmith can be folded into existing engineering workflows instead of operating as a bolt-on dashboard — fitting for its core audience of engineering teams building and maintaining LLM-powered applications and agents.