LangSmith Deployment: Should You Use It?
LangSmith Deployment (formerly LangGraph Platform) is a managed hosting platform for stateful agents. This article answers the question most teams skip: do you actually need it? Then covers the three deployment modes with real cost math, the full CLI lifecycle from dev to production, how to validate with LangSmith Studio, the failure modes specific to managed deployments, and a first-30-days runbook.
Quick Reference
- →LangGraph Platform is now LangSmith Deployment (October 2025) — same runtime, same APIs, integrated into LangSmith tracing and Studio
- →Three deployment modes: Cloud ($0.005/run + $0.0036/min uptime, Plus plan required), Self-Hosted (Lite free up to 1M nodes/month), Hybrid (your data plane + LangSmith control plane, Enterprise only)
- →CLI lifecycle: langgraph dev (no Docker, fast iteration) → langgraph up (Docker parity) → langgraph build (image) → langgraph deploy (push to Cloud)
- →LangSmith Studio (formerly LangGraph Studio) runs in the browser since v2 — no desktop app required; use it to debug individual failures, not as a substitute for automated evals
- →At 10K runs/day on Cloud Plus: ~$50/day in deployment-run fees + ~$5/day uptime — do the math before choosing Cloud over Self-Hosted Lite
- →Self-Hosted Lite is free: deploy the Agent Server yourself, no per-run fees, but you provision and manage PostgreSQL and Redis
- →If your agent completes in one request with no cross-request state, FastAPI + LangGraph is simpler, cheaper, and has no licensing requirements
Should You Use LangSmith Deployment?
Most platform articles skip the most valuable section: when NOT to use the platform. If you skip this and the answer is 'no,' you'll spend 18 minutes learning infrastructure you don't need. The rest of this article assumes you've decided to adopt — so decide here first.
| Use LangSmith Deployment when... | Skip it when... |
|---|---|
| Your agent needs persistent conversation threads that survive across requests | Your agent completes in one request — no cross-request state needed |
| You need human-in-the-loop (HITL) with the ability to resume interrupted runs | Your agent is fully automated with no interrupt or approval workflows |
| You want cron scheduling, background runs, and webhooks without building them | You only need on-demand runs triggered by user requests |
| You don't have a DevOps team to manage agent infrastructure | You already have Kubernetes, a queue, and state management — you're paying for convenience you have |
| You want LangSmith tracing, Studio, and deployment integrated in one platform | Your compliance team won't approve LangChain's data plane for the Cloud or Hybrid modes |
| Your run volume is under ~50K runs/day (Cloud mode stays economical) | Your volume makes $0.005/run more expensive than self-hosting — do the math in Section 2 |
If you skip LangSmith Deployment, you need: a web server (FastAPI/Express), a state store (PostgreSQL or a custom checkpointer), a queue for background runs (Redis/SQS), and your own streaming layer. That's 2–4 weeks of infrastructure work for a team of two. LangSmith Deployment eliminates that — at the cost of vendor lock-in and per-run pricing. The economics favor LangSmith Deployment for teams without existing infra and run volumes below ~50K/day. Above that, self-hosting on your own infrastructure becomes cheaper, even accounting for ops overhead.
Decision tree: which LangSmith Deployment mode fits your constraints
This article covers the deployment decision and lifecycle — which mode to choose, what it costs, how to go from dev to production. The Agent Server Architecture article covers runtime internals: Assistants, Threads, Runs, durability modes, the Go runtime, and failure handling inside the Agent Server. Read this one first.