Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
/**
* Lookups in fixed tables keyed by text the model sends. A plain object also
* answers for the keys it inherits ("constructor", "toString", "__proto__"),
* which would hand a function or a prototype to code expecting an entry.
*/
/** The entry `table` has for `key` itself, or undefined. */
export function own<T>(table: Partial<Record<string, T>>, key: string): T | undefined {
return Object.hasOwn(table, key) ? table[key] : undefined
}
/**
* `value` as one of `allowed`, read without regard to case; `fallback` when it
* is unset, and also when it names none of them, which `warnings` records so
* the model learns its choice was not used.
*/
export function choose<T extends string>(
value: unknown,
allowed: readonly T[],
fallback: T,
field: string,
warnings: string[]
): T {
if (value === undefined || value === null || value === '') return fallback
const text = String(value).trim().toLowerCase()
const hit = allowed.find(name => name.toLowerCase() === text)
if (hit !== undefined) return hit
warnings.push(
`${field} ${JSON.stringify(value)} is not one of ${allowed.join(', ')}; "${fallback}" was used.`
)
return fallback
}