The existing deployment mechanism for agent products outlined in our docs (Agent Builder, Insights) leverages the LangSmith Deployments control plane to coordinate bringup and revisions for agents. This document outlines how to deploy agent products (Agent Builder or Insights) without requiring the control plane for LangSmith Deployments. We leverage standalone agents instead and instruct your LangSmith installation to use them.
You must be on LangSmith version v0.13.19 to follow this doc.
Agent Builder
Use our standalone helm chart (or other method of deployment if available) to bring up Agent Builder.
Here is an example of a values file if using the helm chart. Make sure the environment variables are set.
namespace: <namespace>
images:
apiServerImage:
repository: docker.io/langchain/agent-builder-deep-agent # Or mirrored image path
tag: "<use same tag as langsmith>" # (example: 0.13.18)
config:
# Note that you can read from an existing secret here if needed
langGraphCloudLicenseKey: <license key with agent builder enabled> # Reach out to your rep to confirm agent builder is enabled on your key
# Configure ingress as needed for your setup.
# IMPORTANT: Only requirement is that the hostname must be a subdomain of your langsmith instance
# Example: if langsmith is at "langsmith.mycompany.com", you can make this hostname "agent-builder.langsmith.company.com"
ingress:
enabled: true
hostname: "" # Set this to the hostname you want to use for the ingress.
studioHostname: "" # Set this to the hostname you want to use for the studio ingress
ingressClassName: ""
annotations: {}
labels: {}
tls: []
apiServer:
service:
type: ClusterIP
deployment:
extraEnv:
- name: LANGSMITH_LICENSE_REQUIRED_CLAIMS
value: "agent_builder_enabled"
- name: X_SERVICE_AUTH_JWT_SECRET
value: "<same key salt used in langsmith>"
- name: AGENT_BUILDER_ENCRYPTION_KEY
value: "<same agent builder encryption key used in langsmith>"
- name: SMITH_BACKEND_ENDPOINT
value: "https://<your langsmith host>"
- name: GO_ENDPOINT
value: "https://<your langsmith host>/api/v1"
- name: HOST_BACKEND_ENDPOINT
value: "https://<your langsmith host>/api-host"
# If using the tool server in langsmith, set this as well:
- name: MCP_SERVER_URL
value: "https://<your langsmith host>/mcp"
Run a helm install to bring up the standalone agent builder.
Then, in your LangSmith installation, make sure to set the following:
config:
agentBuilder:
enabled: true
# If you use an existing secret, you will need to put this key there under "agent_builder_encryption_key"
# You can generate the key with this: python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
encryptionKey: <fernet encryption key. should match the standalone agent builder key>
hostBackend:
enabled: true
frontend:
deployment:
extraEnv:
- name: "VITE_AGENT_BUILDER_DEPLOYMENT_URL_PUBLIC"
value: "https://<standalone agent builder hostname>"
- name: "VITE_AGENT_BUILDER_GENERATOR_DEPLOYMENT_URL_PUBLIC"
value: "https://<standalone agent builder hostname>"
# If you want to use the bundled tool server:
- name: "VITE_AGENT_BUILDER_MCP_SERVER_URL"
value: "https://<your langsmith hostname>/mcp"
# If you want to use the bundled trigger server:
- name: "VITE_AGENT_BUILDER_TRIGGERS_API_URL"
value: "https://<your langsmith hostname>"
commonEnv:
- name: "LANGGRAPH_API_URL_PUBLIC"
value: "https://<standalone agent builder hostname>"
# Important: If you are using SSO/Oauth, please make sure the following is also set:
platformBackend:
deployment:
extraEnv:
- name: "OAUTH_COOKIE_DOMAIN"
value: "<langsmith hostname without the protocol>"Run a helm install/upgrade on your LangSmith installation.
You should now see an drop down in your LangSmith frontend (in the top left corner of the page) allowing you to use Agent Builder!
Insights
Use our standalone helm chart (or other method of deployment if available) to bring up the clio agent that powers the Insights feature in LangSmith.
Here is an example of a values file if using the helm chart. Make sure the environment variables are set.
namespace: <namespace>
images:
apiServerImage:
repository: docker.io/langchain/langsmith-clio # Or mirrored image path
tag: "<use same tag as langsmith>" # (example: 0.13.18)
config:
# Note that you can read from an existing secret here if needed
langGraphCloudLicenseKey: <license key>
# Configure ingress as needed for your setup.
# IMPORTANT: Only requirement is that the hostname must be a subdomain of your langsmith instance
# Example: if langsmith is at "langsmith.mycompany.com", you can make this hostname "insights.langsmith.company.com"
ingress:
enabled: true
hostname: "" # Set this to the hostname you want to use for the ingress.
studioHostname: "" # Set this to the hostname you want to use for the studio ingress
ingressClassName: ""
annotations: {}
labels: {}
tls: []
apiServer:
service:
type: ClusterIP
deployment:
extraEnv:
- name: X_SERVICE_AUTH_JWT_SECRET
value: "<same key salt used in langsmith>"
- name: CLIO_ENCRYPTION_KEY
value: "<same insights encryption key used in langsmith>"
Run a helm install to bring up the standalone clio/insights agent.
Then, in your LangSmith installation, make sure to set the following:
config:
insights:
enabled: true
# If you use an existing secret, you will need to put this key there under "insights_encryption_key"
# You can generate the key with this: python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
encryptionKey: <fernet encryption key. should match the standalone insights agent key>
commonEnv:
- name: "CLIO_LANGGRAPH_URL"
value: "https://<standalone insights agent hostname>"
# Important: If you are using SSO/Oauth, please make sure the following is also set:
platformBackend:
deployment:
extraEnv:
- name: "OAUTH_COOKIE_DOMAIN"
value: "<langsmith hostname without the protocol>"Run a helm install/upgrade on your LangSmith installation.
You should now be able to run Insights jobs on tracing projects in LangSmith!