Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
'use client'
export const ALLOWED_WORKLOAD_TYPES = [
'deployment',
'statefulset',
'cronjob',
'job',
'daemonset',
] as const
export type WorkloadType = (typeof ALLOWED_WORKLOAD_TYPES)[number]
export type WorkflowRecipeWorkload = {
id: string
type: WorkloadType
image: string
port?: number
transport?: { type: string; path?: string }
env?: Array<{ name: string; value?: string }>
command?: string[]
args?: string[]
security?: {
runAsUser?: number
runAsGroup?: number
fsGroup?: number
addCapabilities?: string[]
}
resources?: {
requests?: { cpu?: string; memory?: string }
limits?: { cpu?: string; memory?: string }
}
volumeMounts?: Array<{ name: string; mountPath: string }>
volumeClaimTemplates?: Array<{
name: string
storageClass?: string
accessMode?: string
size: string
}>
dependsOn?: string[]
envSecret?: {
name: string
keys: Array<{ secretKey: string; envVar: string }>
}
egressBindings?: Array<{
egressClass?: 'exact-host' | 'public-web'
dns?: string
port?: number
protocol?: 'TCP' | 'UDP'
}>
}
export type WorkflowRecipeStepRun = {
type: 'snippet'
language: 'typescript'
code: string
capabilities?: Record<string, unknown>
}
type WorkflowRecipeStepBase = {
id: string
dependsOn?: string[]
mcpServers?: string[]
agent?: { provider?: string; model?: string }
timeoutSeconds?: number
backoffSeconds?: number
maxIterations?: number
requiresApproval?: {
target: { userId?: string; teamId?: string }
message: string
timeoutSeconds?: number
}
}
export type WorkflowRecipeStep =
| (WorkflowRecipeStepBase & { instruction: string; run?: never })
| (WorkflowRecipeStepBase & { instruction?: never; run: WorkflowRecipeStepRun })
| (WorkflowRecipeStepBase & { instruction?: never; run?: never })
export type WorkflowRecipeMcpServerRef = {
id: string
endpoint: string
}
export type WorkflowRecipeBinding = {
from: string
to: string
port: number
protocol?: 'TCP' | 'UDP'
}
export type WorkflowTriggerOnDemand = {
requiresApproval?: boolean
allowedActors?: Array<'user' | 'autonomous' | 'scheduled'>
}
export type WorkflowTriggerSchedule = {
cron: string
timezone?: string
concurrencyPolicy?: 'Forbid' | 'Replace' | 'Allow'
suspend?: boolean
registeredUserId?: string
}
export type WorkflowTriggers = {
onDemand?: WorkflowTriggerOnDemand
schedule?: WorkflowTriggerSchedule
}
export type WorkflowRunRetention = {
successfulHistoryLimit?: number
failedHistoryLimit?: number
ttlSecondsAfterFinished?: number
maxRunDurationSeconds?: number
}
export type WorkflowRecipeDeclaredResource = {
id: string
type: 'pvc' | 'secret' | 'configmap' | string
data?: Record<string, string>
generateKeys?: string[]
}
export type WorkflowRecipeSpec = {
contextRef?: string
coordinatorImage?: string
workloads?: WorkflowRecipeWorkload[]
runtimeEgress?: {
http?: {
egressClass?: 'exact-host' | 'public-web'
allowedHosts?: string[]
}
}
resources?: WorkflowRecipeDeclaredResource[]
steps?: WorkflowRecipeStep[]
mcpServers?: WorkflowRecipeMcpServerRef[]
agent?: {
provider?: string
model?: string
secretRef?: Record<string, unknown>
soulRef?: Record<string, unknown>
}
// Recipe-level security posture. Matches the CRD schema
// (charts/clerum-crds/crds/workflowrecipe.yaml :: spec.security).
// Typed explicitly so callers like `checkPolicyL1` in RecipeEditor can
// read `spec.security?.allowContextRef` without unsafe `as` casts.
security?: {
isolationLevel?: 'minimal' | 'standard' | 'strict'
allowContextRef?: boolean
}
output?: {
name?: string
destination?: 'configmap' | 'secret' | 'stdout' | 'pvc'
namespace?: string
format?: 'pdf' | 'xlsx' | 'json' | 'text' | 'html' | 'multi'
storageSize?: string
}
inputs?: Record<string, unknown>
inputContract?: Record<string, unknown>
activeProfile?: string
profiles?: Record<string, Record<string, unknown>>
computed?: Array<{ name: string; expression: string }>
networkBindings?: Array<{ workloadId: string; targetContextRef: string }>
bindings?: WorkflowRecipeBinding[]
triggers?: WorkflowTriggers
runRetention?: WorkflowRunRetention
}
export type WorkflowRecipeResource = {
apiVersion?: string
kind?: string
metadata?: { name: string; namespace?: string; creationTimestamp?: string }
spec?: WorkflowRecipeSpec
status?: {
phase?: 'Pending' | 'Running' | 'Succeeded' | 'Failed' | 'Unknown'
workloads?: Array<{ id: string; ready: boolean; replicas?: number }>
message?: string
}
}
export type OperatorDefaults = {
security: {
allowedCapabilities: string[]
maxRunAsUser: number
requireNonRoot: boolean
}
storage: {
defaultStorageClass: string
defaultAccessMode: string
maxPvcSizeGi: number
outputPath: string
}
resources: {
defaultCpuRequest: string
defaultMemoryRequest: string
defaultCpuLimit: string
defaultMemoryLimit: string
}
namespaces: {
mcpWorkloads: string
nonMcpWorkloads: string
}
registry: {
prefix: string
imagePullSecrets: string[]
}
}
export type ValidationIssue = {
phase: 'parse' | 'schema' | 'security'
severity: 'error' | 'warning' | 'info'
path: string
message: string
}
export type ValidationResult = {
valid: boolean
issues: ValidationIssue[]
parsed?: WorkflowRecipeResource
}
export type EnrichmentDiff = {
path: string
before: unknown
after: unknown
reason: string
}
export type EnrichmentResult = {
enriched: WorkflowRecipeSpec
diffs: EnrichmentDiff[]
}