AivexaNewsSearch
AI news for builders and product teamsChecked every hour

LangSmith BYOC on AWS is generally available

Collected Nov 9, 2026

LangChain announced that LangSmith Bring Your Own Cloud (BYOC) is now generally available on Amazon Web Services. The offering gives Enterprise customers a managed LangSmith deployment that runs inside their own AWS account and virtual private cloud, with LangChain provisioning and operating the stack on their behalf. Availability spans 15 AWS regions across the US, EU, and APAC, and organizations enable it by contacting LangChain.

The basic split is between data ownership and operational burden. Traces, datasets, experiments, prompts, agent deployments and sandbox data all remain in the customer's AWS environment, while LangChain handles monitoring, upgrades, scaling and cluster lifecycle management. The pitch is aimed at teams moving agents from pilot to production, where traces and runtime data often carry PII, PHI, customer records, internal API responses and tool outputs, and where keeping that data inside an already governed cloud boundary can otherwise hold up rollout.

The architecture divides into two planes. The control plane runs in LangChain's cloud and covers authentication, user management, organization and workspace configuration, billing and usage metadata, and the LangSmith frontend. It also provisions, monitors and orchestrates customer data planes, and according to LangChain it does not hold sensitive application data. The data plane runs in the customer's AWS account and holds the VPC, a private Amazon EKS cluster, databases, object storage and the sensitive LangSmith data itself, including traces, prompts, datasets, experiments, evaluators, Insights runs, annotation queues, agent deployments and sandboxes.

Communication between the two planes uses AWS PrivateLink rather than the public internet. The management path exposes only the Kubernetes API server of the customer's cluster, which the control plane uses to install and reconcile LangSmith components. The runtime path lets the data plane call back to the control plane for authentication, workspace configuration, telemetry and billing metadata. The EKS cluster is private, with no public API server endpoint and no public IPs on worker nodes.

Each deployment follows a standardized architecture similar to what LangChain uses to run LangSmith Cloud. That includes a dedicated VPC, the private EKS cluster, worker nodes without public IPs, a private load balancer, DNS and ingress on the LangChain BYOC domain, daily backups of Amazon RDS and the trace store written to Amazon S3 in the customer's account, autoscaling for LangSmith services and node capacity, and multi-AZ Amazon RDS and Amazon ElastiCache with automatic failover.

Provisioning relies on Crossplane. Customers apply a provided Terraform module to create the IAM role that LangChain assumes for provisioning; Crossplane then uses that role to create and manage infrastructure. The role is scoped to infrastructure management, and where AWS supports it, permissions are limited to resources carrying a managed_by=langsmith tag and specific name prefixes. Notably, LangChain says the role excludes data-read permissions such as s3:GetObject on the trace bucket, rds-db:connect to PostgreSQL or elasticache:Connect to Redis. In practice, that scope is the detail security reviewers will scrutinize most closely, since the whole premise is that the vendor can operate the platform without reading the data inside it.

Onboarding is intended to be lighter than standing up a self-hosted deployment. LangChain positions plain self-hosting as the option for teams needing full control, but one that brings real operational responsibility: sizing and maintaining databases, running Kubernetes, configuring networking, applying upgrades, monitoring reliability, handling backups and supporting the deployment over time. BYOC aims to give the data residency and network isolation of self-hosting with a managed-service experience. Because the data plane sits in the customer's VPC, agents and sandboxes can reach internal databases, private APIs and on-premises systems over existing network paths, and additional data planes can be created across environments, AWS accounts and regions while organization-level configuration such as users and roles carries across deployments.

Auditability is built into the model. LangChain says urgent support uses explicit, customer-approved access patterns. LangSmith audit logs stay in the data plane, Amazon EKS audit logs go to Amazon CloudWatch in the customer's account, and VPC flow logs land in an Amazon S3 bucket the customer owns. A quoted customer, Rishabh Jain, cofounder of Latent Health, said the setup gives the team the observability tooling it needs while keeping sensitive data in its own secure environment.

Most LangSmith capabilities are available on BYOC at launch. LangChain lists observability (tracing, projects, dashboards, alerts), evaluation (datasets, experiments, evaluators, annotation queues), Insights for automatic analysis of traces to surface usage patterns and failure modes, Context Hub for prompts and other context files, LangSmith Deployment for running agents inside the customer data plane, Sandboxes for code execution and filesystem work on a dedicated node group, LLM Gateway for model access through one API key with centralized spend and rate-limit policies, Fleet for no-code agents, and SmithDB persisting trace data to Amazon S3 in the customer account. Managed Deep Agents, the LLM auth proxy and Engine also support BYOC. The word "most" is worth noting, since it implies some features remain cloud-only.

This is a competitive positioning move as much as a technical one. Enterprise agent platforms increasingly compete on data governance, and the BYOC pattern, where the vendor runs the control plane and the customer hosts the data plane, is a common answer for regulated buyers who will not send traces to a third-party SaaS. A likely trade-off is that customers take on cloud spend and quota management for infrastructure they own, even though LangChain operates it, and that regional availability is limited to the 15 listed regions.

Why it matters: Teams in financial services, healthcare, cybersecurity and other regulated environments that have been blocked from adopting LangSmith because traces contain sensitive data now have a managed option that keeps that data in their own AWS account. For developers, it means tracing, evaluation and deployment can run against private systems without a separate operating model; for platform teams, the Terraform-based setup and narrowly scoped IAM role lower the review burden, though verifying that scope is likely to be part of any security assessment. Competitors offering similar bring-your-own-cloud arrangements can expect buyers to compare the permission boundaries and feature parity directly.

Read at LangChain

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

LangSmith Bring Your Own Cloud is now generally available on AWS, giving Enterprise teams managed observability, evaluation, and deployment inside their own VPC.