Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import type { HostAllowedModel, LlmModelCatalogEntry, LlmPolicy, LlmProvider } from '@/lib/llm'
// Write-only credential wiring. Values are NEVER read back from the server; the
// draft holds only what the operator typed this session. `existingKeys` lights
// the present/absent chips in edit mode without exposing any value, and a blank
// field means "leave the stored key unchanged" (merge PUT on save).
export type LlmCredentialWiring = {
draft: Record<string, string>
onChange: (dataKey: string, value: string) => void
// Keys already stored in the Host's Secret (edit mode). Create omits it.
existingKeys?: string[]
}
export type LlmProviderConfigProps = {
// PRIMARY model (spec.model). Always required.
provider: LlmProvider
model: string
onPrimaryChange: (next: { provider: LlmProvider; model: string }) => void
// Fallback policy (spec.llmPolicy). `undefined` = the Host has no fallback.
policy: LlmPolicy | undefined
onPolicyChange: (next: LlmPolicy | undefined) => void
// Per-host model allowlist subset (spec.allowedModels, Topic 3a). The flat
// (provider, model) pairs = WHICH models this host offers its end users. Empty
// = unrestricted (offer the full global allowlist per provider). Controlled by
// the parent, which assembles/omits it in the Host spec.
allowedModels: HostAllowedModel[]
onAllowedModelsChange: (next: HostAllowedModel[]) => void
// Operator model allowlist (enabled + disabled rows); pickers show the enabled
// subset per provider.
catalog: LlmModelCatalogEntry[]
catalogLoading?: boolean
catalogError?: string
// Copy and model-restriction controls vary by surface. The agent creation
// flow selects one default model without exposing per-host restrictions,
// while the edit surface retains the full allowed-models editor.
modelLabel?: string
showAllowedModels?: boolean
// Existing-agent layout: preserve the saved default model internally and
// put the primary provider's Allowed models control in the model field's
// place. Provider changes still resolve a valid default model automatically.
replacePrimaryModelWithAllowedModels?: boolean
// Credential editing. Omit to render the model + fallback structure WITHOUT
// credential inputs (create flow reusing an existing shared Secret, where the
// wizard cannot introspect the Secret's keys).
credentials?: LlmCredentialWiring
// Data keys already present in the Host's own Secret — feeds the fallback
// credentialSlot dropdown's extra keys (e.g. `claude-api-key-fb1`).
secretKeys?: string[]
// Creation keeps the optional fallback configuration out of the way until an
// operator explicitly opens it. Existing-agent editing stays expanded.
fallbackProvidersInitiallyCollapsed?: boolean
disabled?: boolean
/** Proven by a successful Grok connections list. Hidden when false/omitted. */
grokEnabled?: boolean
}