spec.deployment

Where and how the agent ships in production. Pick a target with type, then configure the matching sub-block.

spec:
  deployment:
    type: kubernetes
    kubernetes:
      image:
        registry: ghcr.io
        repository: company/customer-support-agent
        tag: v1.0.0

Fields

Field	Type	Required	Description
type	string		enum: kubernetes, cloudrun, vercel, cloudflare. Selects the deployment target.
kubernetes	object		See Kubernetes.
cloudrun	object		See Cloud Run.
vercel	object		See Vercel.
cloudflare	object		See Cloudflare.

Kubernetes

spec:
  deployment:
    type: kubernetes
    kubernetes:
      image:
        registry: ghcr.io
        repository: company/customer-support-agent
        tag: v1.0.0

Field	Description
image	See Image.

The Kubernetes block is intentionally minimal in v1 - it carries the image and leaves the rest (Deployment, Service, ConfigMap, …) to the generator's templating. Future minor versions may add typed slots for replica counts and resource limits as common patterns harden.

Cloud Run

spec:
  deployment:
    type: cloudrun
    cloudrun:
      image:
        registry: us-central1-docker.pkg.dev
        repository: my-project/agents/customer-support
        tag: v1.0.0
        useCloudBuild: true
      resources:
        cpu: "1"
        memory: 512Mi
      scaling:
        minInstances: 1
        maxInstances: 10
        concurrency: 80
      service:
        timeout: 300
        allowUnauthenticated: false
        serviceAccount: [email protected]
        executionEnvironment: gen2
      environment:
        LOG_LEVEL: info
        CACHE_TTL: "300"

Field	Reference
image	Image
resources	Resources
scaling	Scaling
service	Service
environment	{ [key: string]: string } - env vars.

environment is injected into the running service and commonly carries both configuration and secrets. Never inline a real secret here - reference it with a ${VAR} placeholder and let your deploy pipeline supply the value. See Secrets & interpolation. (The kubernetes target has no environment map of its own; the generator's templating owns the Deployment's env / envFrom.)

Resources

Field	Type	Description
cpu	string	CPU allocation per instance (e.g. "1").
memory	string	Memory per instance (e.g. 512Mi, 2Gi).

Scaling

Field	Type	Description
minInstances	integer	Minimum number of instances kept warm.
maxInstances	integer	Cap on instances Cloud Run may spin up.
concurrency	integer	Concurrent requests served per instance.

Service

Field	Type	Description
timeout	integer	Per-request timeout in seconds.
allowUnauthenticated	boolean	If true, the service is publicly invokable.
serviceAccount	string	GCP service-account email the service runs as.
executionEnvironment	string	Cloud Run execution environment (e.g. gen1, gen2).

Vercel

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

Field	Reference
project	string - Vercel project name.
team	string - Vercel team ID or slug the project belongs to.
framework	string - Framework identifier (e.g. nextjs). Omit for auto-detection.
runtime	string - enum: nodejs, edge. Vercel function runtime.
regions	string[] - Region identifiers (e.g. iad1).
functions	object - Functions configuration.
environment	{ [key: string]: string } - Environment variables.

Unlike kubernetes and cloudrun which deploy a prebuilt container image, Vercel deploys from source via its own build pipeline. There is no image sub-block – the build is managed by Vercel based on the project's framework and root directory.

environment is injected at deploy time and commonly carries both configuration and secrets. Never inline a real secret here - reference it with a ${VAR} placeholder and let your deploy pipeline supply the value. See Secrets & interpolation.

Functions

Field	Type	Description
memory	integer	Memory limit per function invocation in MB (e.g. 1024).
maxDuration	integer	Maximum execution time in seconds.

Cloudflare

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

Field	Reference
name	string - Worker (script) name registered with Cloudflare.
accountId	string - Cloudflare account ID. Prefer a ${VAR} placeholder.
compatibilityDate	string - Workers runtime compatibility date (e.g. 2025-01-01).
compatibilityFlags	string[] - Runtime compatibility flags (e.g. nodejs_compat).
routes	string[] - Custom routes / domains (e.g. agent.example.com/*).
workersDev	boolean - Expose the Worker on its *.workers.dev subdomain.
environment	{ [key: string]: string } - Plain-text vars (wrangler vars).

Like vercel and unlike kubernetes/cloudrun, Cloudflare Workers deploy from source via wrangler rather than a prebuilt container image, so there is no image sub-block. This target models Workers (the server/serverless product, the right fit for an A2A agent server), not Pages. Workers always run on the V8-isolate edge runtime, so there is no runtime enum as on Vercel - Node.js API needs are met with the nodejs_compat compatibility flag. compatibilityDate is effectively required by wrangler; when omitted the generator supplies a default.

environment carries plain-text wrangler vars. Never inline a real secret here - reference it with a ${VAR} placeholder, and set true secrets out-of-band with wrangler secret put. See Secrets & interpolation.

Image

Used by the kubernetes and cloudrun sub-blocks only.

Field	Type	Description
registry	string	Image registry (e.g. ghcr.io).
repository	string	Image repository within the registry.
tag	string	Image tag.
useCloudBuild	boolean	If true, the generator wires in Google Cloud Build for image production. Most useful with type: cloudrun.