I’ve been working with LangGraph agents for some company projects and I’m running into deployment challenges. The documentation seems to focus mainly on using LangSmith for production deployments, but I need to host everything on our own infrastructure.
I want to deploy these agents on our internal servers without depending on LangSmith’s API or services. The main issue I’m facing is finding stable deployment methods that don’t require external platform integration.
Has anyone successfully deployed LangGraph agents in a self-contained environment? What are the best practices for running them independently without third-party dependencies? I’m looking for production-grade solutions rather than just development setups.
Had this exact problem six months ago - our compliance team shut down any external dependencies too. I ended up throwing the whole LangGraph app into Docker containers and running it on our Kubernetes cluster. Here’s the thing: LangGraph doesn’t actually need LangSmith to work, they just push it hard in the docs. You can totally run it standalone if you handle state persistence and monitoring yourself. I swapped in PostgreSQL for state management and built custom logging instead of using LangSmith’s tracing. Watch out for memory issues in production - those long agent conversations pile up state fast. Also set up your own health checks and restart logic since you won’t have LangSmith’s platform monitoring. Works great for us though - we’re hitting 500 agent interactions daily with zero problems.