Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
/**
* Bounds applied to a subscription provider's discovered model catalog before
* it is reconciled into `codex_catalog_models` / `grok_catalog_models`. The
* upstream list is provider-controlled; without a bound one sync could insert
* an unbounded number of enabled rows (and republish them in the runtime
* allowlist ConfigMap). Shared by the Codex and Grok catalog services so both
* brokers enforce the same limits (the proxies mirror them).
*/
export const MAX_DISCOVERED_CATALOG_MODELS = 256
export const MAX_CATALOG_MODEL_ID_LENGTH = 128
export type BoundedDiscoveredCatalogModels<T> = {
models: T[]
/** Entries whose id is not a non-empty string of at most 128 characters. */
droppedInvalidId: number
/** Entries whose (valid) id falls beyond the first 256 distinct ids. */
droppedOverCount: number
}
/**
* Keeps entries with a valid model id whose id is among the first
* `MAX_DISCOVERED_CATALOG_MODELS` distinct ids, in provider order. Duplicate
* ids of a kept model stay in the list so reconcile keeps its existing
* last-entry-wins metadata semantics.
*/
export function boundDiscoveredCatalogModels<T extends { model: unknown }>(
models: readonly T[]
): BoundedDiscoveredCatalogModels<T> {
const kept: T[] = []
const distinct = new Set<string>()
let droppedInvalidId = 0
let droppedOverCount = 0
for (const entry of models) {
const id = entry?.model
if (typeof id !== 'string' || id.length === 0 || id.length > MAX_CATALOG_MODEL_ID_LENGTH) {
droppedInvalidId += 1
continue
}
if (!distinct.has(id)) {
if (distinct.size >= MAX_DISCOVERED_CATALOG_MODELS) {
droppedOverCount += 1
continue
}
distinct.add(id)
}
kept.push(entry)
}
return { models: kept, droppedInvalidId, droppedOverCount }
}