Deployment Targets

spec.deployment tells the generator where the agent ships. You pick one target with type and fill in the matching sub-block. All supported targets are shown below as complete, copy-pasteable manifests for the same agent.

Kubernetes

The Kubernetes block is intentionally minimal in v1: it carries the container image and leaves the Deployment, Service, and ConfigMap to the generator's templating.

apiVersion: adl.inference-gateway.com/v1
kind: Agent
metadata:
  name: analytics-agent
  description: Analytics agent deployed to Kubernetes from GHCR
  version: "2.1.0"
spec:
  capabilities:
    streaming: true
    pushNotifications: false
    stateTransitionHistory: false

  agent:
    provider: anthropic
    model: claude-sonnet-4-7

  server:
    port: 8080
    scheme: https

  language:
    go:
      module: github.com/example/analytics-agent
      version: "1.26"

  deployment:
    type: kubernetes
    kubernetes:
      image:
        registry: ghcr.io
        repository: example/analytics-agent
        tag: v2.1.0

Cloud Run

The same agent, retargeted at Cloud Run. The cloudrun block adds typed slots for resources, autoscaling, and service settings, and can opt into Google Cloud Build with useCloudBuild: true.

apiVersion: adl.inference-gateway.com/v1
kind: Agent
metadata:
  name: analytics-agent
  description: Analytics agent deployed to Cloud Run with Cloud Build
  version: "2.1.0"
spec:
  capabilities:
    streaming: true
    pushNotifications: false
    stateTransitionHistory: false

  agent:
    provider: anthropic
    model: claude-sonnet-4-7

  server:
    port: 8080
    scheme: https

  language:
    go:
      module: github.com/example/analytics-agent
      version: "1.26"

  deployment:
    type: cloudrun
    cloudrun:
      image:
        registry: us-central1-docker.pkg.dev
        repository: my-project/agents/analytics-agent
        tag: v2.1.0
        useCloudBuild: true
      resources:
        cpu: "2"
        memory: 1Gi
      scaling:
        minInstances: 0
        maxInstances: 20
        concurrency: 40
      service:
        timeout: 600
        allowUnauthenticated: false
        serviceAccount: [email protected]
        executionEnvironment: gen2
      environment:
        LOG_LEVEL: info
        CACHE_TTL: "300"

Vercel

The same agent, retargeted at Vercel. Unlike kubernetes and cloudrun, Vercel deploys from source via its own build pipeline -- there is no prebuilt container image.

apiVersion: adl.inference-gateway.com/v1
kind: Agent
metadata:
  name: analytics-agent
  description: Analytics agent deployed to Vercel
  version: "2.1.0"
spec:
  capabilities:
    streaming: true
    pushNotifications: false
    stateTransitionHistory: false

  agent:
    provider: anthropic
    model: claude-sonnet-4-7

  server:
    port: 8080
    scheme: https

  language:
    typescript:
      packageName: analytics-agent
      nodeVersion: "20"

  deployment:
    type: vercel
    vercel:
      project: analytics-agent
      team: acme
      runtime: nodejs
      regions: [iad1]
      functions:
        memory: 1024
        maxDuration: 60
      environment:
        LOG_LEVEL: info

Cloudflare

The same agent, retargeted at Cloudflare Workers. Like vercel and unlike kubernetes/cloudrun, Workers deploy from source via wrangler -- there is no prebuilt container image. Workers run on the V8-isolate edge runtime, so Node.js APIs are enabled with the nodejs_compat compatibility flag rather than a runtime choice.

apiVersion: adl.inference-gateway.com/v1
kind: Agent
metadata:
  name: analytics-agent
  description: Analytics agent deployed to Cloudflare Workers
  version: "2.1.0"
spec:
  capabilities:
    streaming: true
    pushNotifications: false
    stateTransitionHistory: false

  agent:
    provider: anthropic
    model: claude-sonnet-4-7

  server:
    port: 8080
    scheme: https

  language:
    typescript:
      packageName: analytics-agent
      nodeVersion: "20"

  deployment:
    type: cloudflare
    cloudflare:
      name: analytics-agent
      accountId: ${CLOUDFLARE_ACCOUNT_ID}
      compatibilityDate: "2025-01-01"
      compatibilityFlags:
        - nodejs_compat
      routes:
        - analytics.example.com/*
      workersDev: false
      environment:
        LOG_LEVEL: info

Highlights

• type selects one target. Set it to kubernetes, cloudrun, vercel, or cloudflare; the generator reads the matching sub-block and ignores the others.
• kubernetes and cloudrun share the image shape. Both targets reuse the same image object (registry, repository, tag, useCloudBuild).
• Vercel and Cloudflare deploy from source, not from an image. Neither has an image sub-block; Vercel manages its own build pipeline, and Cloudflare Workers build and ship via wrangler. Workers run on the V8-isolate edge runtime, so Node.js APIs come from the nodejs_compat compatibility flag rather than a runtime choice.
• Cloud Run carries more knobs. resources, scaling, and service give Cloud Run typed control over CPU/memory, instance counts, and request handling. scaling.minInstances: 0 lets it scale to zero when idle.
• Kubernetes stays lean on purpose. v1 keeps it to the image so the generator's templates own replica counts and limits. Future minor versions may add typed slots as patterns harden.