> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.fabricgateway.ai/run/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.fabricgateway.ai/_mcp/server. # Run the data plane in your VPC > What the Fabric AI Gateway data plane is, what you run versus what Postman runs, what the Helm chart installs, and how the data plane connects to Postman. This section applies to the Enterprise plan, where the control plane runs in Postman cloud and you run the data plane in your own VPC. The data plane is the part of Fabric AI Gateway that carries request traffic. You install it with a Helm chart in your own Kubernetes cluster, next to the systems it calls, and it stays enrolled with Postman's control plane. This page explains who runs what, what the chart installs, and how the data plane connects to Postman. ## How the two planes fit together * **The control plane** is hosted and managed by Postman. It's where you define connections, routes, services, plugins, and policy, in the console or through the admin API. * **The data plane** runs in your cluster. When a gateway pod starts, it enrolls with the control plane, pulls down your configuration, and serves requests with it. At runtime, a client sends a request to the gateway. The gateway applies the routing and policy it synced from the control plane, adds upstream credentials where needed, and forwards the request to the LLM provider, MCP server, or API. Telemetry goes to whatever backend you point the data plane at. Your configuration isn't in the Helm chart. Connections, routes, and plugins arrive over the sync connection, so you change them in the console, not with `helm upgrade`. ## What Postman runs and what you run | Postman provides and runs | You provide and run | | ------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | | The control plane and the console | The Kubernetes cluster, plus `kubectl` and Helm 3 or later | | The Helm chart and the gateway container image | A container registry you mirror the image into | | Your tenant ID and the control plane's sync address | A MySQL database for the data plane, unless you use the distributed secret store | | The control plane's CA certificate, also downloadable from the console | The Kubernetes Secret that holds the encryption key and passwords | | An enrollment credential, either registered in the console (workload attestation) or minted there (bootstrap token) | A telemetry backend, if you want one | | | Ingress, DNS, and TLS for the clients that call the gateway | The data plane's encryption key isn't from Postman. You generate it, and it doesn't need to match anything on the control plane. The chart's install notes say this key must match the control plane's. That note is out of date: generate your own key. See [Secrets and keys](/run/secrets-and-keys). ## What the chart installs With the default values, the chart creates the following in one namespace: | Component | What it does | Default | | -------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------ | | Gateway | A Deployment of gateway pods that carry request traffic. Each pod also runs an OpenTelemetry Collector sidecar that ships traces, metrics, and logs. | 2 replicas | | Gateway Service | A `ClusterIP` Service on port `8080`. | On | | Embedder | A text-embeddings model server that powers semantic search over MCP tools. | 1 replica | | Schema migration Job | Applies database schema changes to MySQL before the gateway rolls out, on every install and upgrade. | Runs to completion | | PodDisruptionBudgets | Keep at least one gateway pod and one embedder pod available during voluntary disruptions. | On | | ServiceAccount | The identity of the gateway pods, also used for workload attestation. | On | | ConfigMaps | The gateway's environment variables and the collector's pipeline. | On | Autoscalers, network policies, a policy (OPA) sidecar, persistent model caching, and the distributed secret store are optional and off by default. A default install uses only core Kubernetes resources and needs no CRDs. ### Names you'll see Most resource names come from the chart, not from your release name, so they're the same whatever you call the release. Because of this, install one release per namespace. | Thing | Name | | -------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ | | Gateway Deployment (a StatefulSet with the distributed secret store), Service, and container | `fabric-gateway-chart` | | Gateway PodDisruptionBudget | `fabric-gateway-chart-pdb` | | Schema migration Job | `fabric-gateway-chart-dp-migrate` | | App Secret (default name) | `fabric-gateway-chart-ext-secret` | | Control plane CA | `fabric-gateway-chart-sync-ca` ConfigMap, or a `fabric-dp-sync-ca` Secret you create | | Embedder Deployment and Service | `fabric-gateway-embedder` | | ServiceAccount | `-sa`, for example `fabric-dp-sa` | Gateway pods carry the label `app.kubernetes.io/name=fabric-gateway-chart`. Embedder pods carry a different label, `app=fabric-gateway-embedder`, so a selector for one never matches the other. These pages use the release name `fabric-dp` and the namespace `pm-fabric-dp`. You can choose another namespace. ## How the data plane connects to Postman The data plane does nothing useful until it has joined the control plane. Every connection is outbound from your cluster. The control plane never connects in to your VPC. With workload attestation, the control plane also reads your cluster's public signing keys from its OIDC issuer URL, unless you paste the keys when you register the trust anchor. If your issuer isn't reachable from the internet, select **Paste JWKS**. See [Connect to the control plane](/run/connect-a-data-plane). 1. **Enroll.** A new gateway pod proves which tenant it belongs to, with its Kubernetes service account token (workload attestation) or with a bootstrap token. It sends the request to the sync address over TLS, which it verifies against the control plane CA certificate you supplied. 2. **Get a certificate.** The control plane issues a short-lived client certificate for that tenant and node. The pod keeps it under `/data/cache` and uses it for mutual TLS from then on. 3. **Sync.** The pod opens a long-lived sync connection and receives your configuration and policy. It keeps the connection open for updates and renews its certificate before it expires. 4. **Serve.** The pod serves client requests with the configuration it received. It starts listening as soon as it's enrolled, so a request that arrives before its first configuration can get a `503` or a `404`. If enrollment fails, the pod never becomes ready. A pod stuck in `CrashLoopBackOff` right after install usually has an enrollment problem, such as the wrong tenant ID, sync address, or CA certificate, not a Kubernetes one. See [Connect to the control plane](/run/connect-a-data-plane). ## Hand the install to an AI agent In a staging or throwaway cluster, an AI agent can follow the install for you. Don't give an agent access to a production cluster until you're comfortable with the steps. Keep secrets in environment variables or your secrets manager, and don't paste them into the prompt. #### Install the data plane in staging Read [https://docs.fabricgateway.ai/get-started/quickstart-data-plane.md](https://docs.fabricgateway.ai/get-started/quickstart-data-plane.md) and [https://docs.fabricgateway.ai/run/install.md](https://docs.fabricgateway.ai/run/install.md). Install the data plane in my staging cluster, using the kubectl context I name. Before each step, show me the command. After each step, run the check that the page gives and tell me whether the result matches. Stop before `helm install` and ask me to approve the staging database migration, or set `migrations.enabled: false` and tell me that migrations were skipped. Don't change any production resources. If a check fails, stop and tell me what you saw. ## Next steps * To check what your cluster needs, see [Requirements and sizing](/run/plan-a-deployment). * To open the right firewall paths, see [Network and egress](/run/network). * To install, see [Install with Helm](/run/install). ## Docs - [Requirements and sizing](https://docs.fabricgateway.ai/run/plan-a-deployment.md): Check the Kubernetes, capacity, and external dependencies the Fabric AI Gateway data plane needs before you install it in your VPC. - [Network and egress](https://docs.fabricgateway.ai/run/network.md): Every network path the Fabric AI Gateway data plane uses, the egress you need to allow, and the ports inside the cluster. - [Install the data plane with Helm](https://docs.fabricgateway.ai/run/install.md): Install the Fabric AI Gateway data plane in your Kubernetes cluster with Helm, see what Helm does in order, and verify the install. - [Database migrations](https://docs.fabricgateway.ai/run/database-migrations.md): How the data plane chart migrates its MySQL schema, and how to adopt an existing database, run migrations yourself, or use a separate migration user. - [Connect the data plane to the control plane](https://docs.fabricgateway.ai/run/connect-a-data-plane.md): Enroll the data plane with Kubernetes workload attestation or a bootstrap token, supply the control plane CA certificate, and check the data plane in the console. - [Expose the gateway to your clients](https://docs.fabricgateway.ai/run/expose-the-gateway.md): Make the Fabric AI Gateway data plane reachable from outside the cluster with an ingress or load balancer, end TLS, and set its host name in your values and the console. - [Data plane production checklist](https://docs.fabricgateway.ai/run/production-checklist.md): Check your data plane's security, availability, and observability settings against the chart's production defaults before it carries production traffic. - [Helm values](https://docs.fabricgateway.ai/run/helm-values.md): Reference for the data plane Helm chart values, environment variable layering, feature toggles, and the defaults to retarget before you install. - [Secrets and keys](https://docs.fabricgateway.ai/run/secrets-and-keys.md): Create the Kubernetes Secret the data plane reads, generate its encryption key, supply the control plane CA certificate, and rotate each value. - [Monitor the data plane](https://docs.fabricgateway.ai/run/monitor.md): Check data plane health with kubectl, its health endpoints, and the console, and collect its logs, metrics, and traces in your own observability stack. - [Scale and run a highly available data plane](https://docs.fabricgateway.ai/run/scale-and-high-availability.md): Run several gateway replicas, add autoscaling, keep node drains working, choose a secret store, and understand what the data plane does when the control plane is unreachable. - [Run the tool-search embedder](https://docs.fabricgateway.ai/run/embedder.md): Understand the embedder the data plane chart deploys for MCP tool search, and keep its model cache, scaling, drains, and network access under control. - [Troubleshoot the data plane](https://docs.fabricgateway.ai/run/troubleshooting.md): Find the cause and fix for common data plane problems, from failed migrations and crashing pods to enrollment errors, missing telemetry, and embedder issues. - [Maintain the data plane and its data](https://docs.fabricgateway.ai/run/maintenance.md): Learn what state the data plane keeps and where, what you need to back up, how long stored data lasts, and the routine tasks that keep a data plane healthy. - [Upgrade the data plane](https://docs.fabricgateway.ai/run/upgrade.md): Upgrade the data plane's chart and image with Helm, understand what runs during an upgrade, and roll back if you need to. - [Data plane chart release notes](https://docs.fabricgateway.ai/run/release-notes.md): See what changed in each version of the data plane Helm chart, and what to do when you upgrade to it. - [Uninstall the data plane](https://docs.fabricgateway.ai/run/uninstall.md): Remove the data plane Helm release, and clean up the Secrets, volumes, database, and console nodes that the uninstall leaves behind.