Choose LangSmith Deployment for LangChain to run the Agent Server, checkpoint database, streaming and scaling; use the open-source library in your own service if your team owns those parts. GILD calculates managed deployment from about US$52 a month for a Serverless Small left running to US$1,536 for Dedicated Large, and database disk storage is not billed. Customer-facing agents belong on Dedicated.
When you need this
- An agent is ready for users, and the team must decide between managed hosting and its own service.
- Your organization budgeted LangGraph Platform on per-run and uptime pricing, which ended on October 1, 2026.
- Security requires agents and their checkpoints to stay inside your network.
Evaluation criteria
Managed or your own service
Who runs the server, the checkpoint database, and the workers?
Look forCompare what each hosting option leaves your team to run. LangSmith Deployment manages threads, checkpoints, streaming, cron, and autoscaling on Agent Server. Your own FastAPI or Node service must implement the API, streaming, cancellation and scaling. After a year on the platform, one team valued managed streaming most. Etsy hosted its graph on a new Kubernetes cluster and streamed through its existing web cluster to reuse authentication.
Plan
Which LangSmith plan does hosting need?
Look forCloud deployments need Plus, US$39 per seat per month, which includes 1 free Small Serverless deployment. The Developer plan has no deployments. Hybrid, and self-hosted with a control plane, require Enterprise.
Deployment type
Is the agent customer-facing?
Look forUse Serverless for background, internal, or test agents. Its scale-to-zero is in beta, and the first request after an idle period waits for startup. Dedicated stays on with its own database, backups, and high availability, and LangChain recommends it for customer-facing agents. You can change a deployment's size later, but its type is fixed at creation.
Monthly compute
What does each size cost for a full month?
Look forGILD priced each size for a full 720-hour month at US$1 per LSU, before autoscaling, from LangChain's hourly size table checked October 28, 2026; database disk storage is not billed. Serverless, if never idle, costs about US$52 for Small, US$220 for Medium and US$441 for Large. Dedicated costs about US$384 for Small, US$768 for Medium and US$1,536 for Large. The runtime bills 0.0675 LSU per vCPU-hour and 0.009 per GiB-hour, and the Dedicated database bills 0.177 per vCPU-hour and 0.025 per GiB-hour.
Several agents per deployment
Do you need one deployment per agent?
Look forNo. Billing follows resources, so several graphs can share one deployment through langgraph.json, sharing its compute, database, scaling, and revision. Give an agent its own deployment only when it needs a different size or isolation.
Network and data
Must agents and checkpoints stay in your network?
Look forHybrid keeps Agent Servers and their data in your VPC under LangChain's control plane. A standalone Agent Server runs on Docker or Kubernetes with your own Postgres and Redis and a LangSmith license key, which the server checks at start-up through beacon.langchain.com unless it runs air-gapped. Cloud deployments run in US or EU data regions and reject payloads over 25 MB.
Tracing
Should run data reach LangSmith?
Look forDecide tracing apart from hosting. Set LANGSMITH_TRACING=false for agents whose data must stay in your systems.
Questions to ask vendors
- Is each agent customer-facing and so Dedicated, or background work that can scale to zero?
- Which size does a load test show, and what is that per month in LSUs?
- Can several of our graphs share one deployment?
- If we run the library ourselves, who owns streaming, cancellation, and the checkpoint database?
- Must Agent Servers run in our VPC, and does that require Enterprise?
Red flags
- Budgets still built on per-run and uptime pricing.
- A customer-facing agent on Serverless with a cold start in front of the first request.
- Dedicated deployments left running for development branches.
- The langgraph dev in-memory server used as a production backend.
- Self-hosting chosen without an owner for Postgres backups and upgrades.
Products in this piece
Sources
- Official documentation
- 10
- Public source
- 2
- LangSmith plans and pricing (LangSmith Deployment rates)
- LangSmith docs: LangSmith Deployment
- LangSmith docs: Cloud platform features (deployment types and sizes)
- LangSmith docs: Manage billing (LangSmith Deployment billing)
- LangSmith docs: Hybrid
- LangSmith docs: Standalone Agent Server
- LangSmith docs: LangGraph CLI
- LangGraph docs: Run a local server
- LangGraph docs: Deployment
- LangChain blog: LangGraph Platform generally available, renamed LangSmith Deployment in Oct 2025 (May 14, 2025)
- Hacker News comment on a year of LangGraph Platform, now LangSmith Deployment (Oct 21, 2025)
- Derrick Kondo (Etsy): The Etsy Gifting Assistant, From Prototype to Production, Interrupt 26 (Jun 8, 2026)