Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import type { HostRuntimeStatus } from '../../../../src/types'
import { Pill } from '../Common'
export interface FallbackBadgeProps {
/** `StatusResponse.servedBy` projected onto the Host runtime status (R5.9). */
servedBy?: HostRuntimeStatus['servedBy']
}
/**
* "Running on fallback" indicator for the chat indicators row (spec §3-R5.9),
* sat next to the ContextWindowIndicator / ModelSelector in AgentWorkspace.
*
* Renders ONLY when the Host is actively serving from a configured fallback
* (`servedBy.fallback === true`). Absent `servedBy` (older mcp-host builds, or a
* Host with no `spec.llmPolicy`) and the normal case (`fallback === false`,
* primary serving) render nothing — no noise. The label names the pair actually
* serving the request so the operator can see the (possibly costlier) fallback.
*/
export function FallbackBadge({ servedBy }: FallbackBadgeProps) {
if (!servedBy || !servedBy.fallback) return null
const pair = `${servedBy.provider}/${servedBy.name}`
const label = `Running on fallback: ${pair}`
const tooltip = `The primary model is unavailable — this agent is temporarily served by a configured fallback (${pair}). It returns to the primary automatically once it recovers.`
return (
<Pill
tone="warning"
size="sm"
className="fallback-badge"
role="status"
aria-label={label}
title={tooltip}
>
<span className="fallback-badge-glyph" aria-hidden="true" />
<span className="fallback-badge-label">{label}</span>
</Pill>
)
}