What causes "AI search request failed" in RocketChat/Rocket.Chat?

Read the err from the log entry — it distinguishes provider auth/rate-limit errors from OpenSearch/index errors

AI search request failed

Error message

AI search request failed

What it means

After AISearch.status() reported intelligent search available, the actual intelligent-search request itself threw. The endpoint logs this warning, resets intelligentResults to an empty array, and still returns HTTP success with the non-intelligent results plus meta flags — so the AI portion fails silently for the caller unless it checks the intelligent array and meta fields.

Source

Thrown at apps/meteor/server/api/v1/ai-search.ts:289

			aiSearchStatus.intelligentSearchConfigured
		) {
			try {
				intelligentResults = await AISearch.search({
					query,
					userId: this.userId,
					filters: {
						rid,
						rids,
						roomNames,
						fromUsername,
						fromUsernames,
						startDate: startDate?.toISOString(),
						endDate: endDate?.toISOString(),
					},
					limit: intelligentLimit,
				});
			} catch (error) {
				this.logger.warn({ msg: 'AI search request failed', err: error });
				intelligentResults = [];
			}
		}

		return API.v1.success({
			intelligent: intelligentResults,
			meta: {
				intelligentSearchEnabled: aiSearchStatus.intelligentSearchEnabled,
				intelligentSearchConfigured: aiSearchStatus.intelligentSearchConfigured,
				answerGenerationConfigured: aiSearchStatus.answerGenerationConfigured,
			},
		});
	},
);

API.v1.get(
	'ai.llm.models',
	{

View on GitHub (pinned to b2c16d5842)

Solutions

Read the err from the log entry — it distinguishes provider auth/rate-limit errors from OpenSearch/index errors
Check AI provider credentials and quota if the error is 401/429 from the provider
Check search backend health and that the workspace index exists and is not mid-rebuild
Narrow the query (fewer rooms, smaller date window) to rule out timeouts
In clients, treat an empty intelligent array as degraded output instead of 'no results'

Defensive patterns

Strategy: fallback

Validate before calling

// server-side guard: only call the intelligent pipeline when status is green
if (!aiSearchStatus.hasIntelligentSearchLicense || !aiSearchStatus.intelligentSearchEnabled || !aiSearchStatus.intelligentSearchConfigured) {
	return API.v1.success({ intelligent: [], meta: { ...aiSearchStatus } });
}

Try / catch

let intelligentResults: AISearchResult[] = [];
try {
	intelligentResults = await runIntelligentSearch(query, { limit: intelligentLimit });
} catch (error) {
	this.logger.warn({ msg: 'AI search request failed', err: error });
	intelligentResults = []; // degrade to normal search, still return success
}

Prevention

Monitor and alert on the warn's err field to catch expired AI provider keys before users notice
Keep the search index healthy; reindex during maintenance windows
Include the meta flags in client telemetry so silent degradation is measurable

When it happens

Trigger: The search pipeline (OpenSearch/vector store) erroring on this specific query: missing or rebuilding index, embedding/LLM provider returning 4xx/5xx or timing out, invalid combinations of filters (rid/roomNames/fromUsername/date ranges) the backend rejects, or result size beyond limits.

Common situations: OpenSearch index being reindexed; AI provider API key expired or rate-limited; very wide date ranges or many rooms making the query slow enough to hit timeouts; status() passing but the query path misconfigured.

Related errors

AI-assisted analysis of RocketChat/Rocket.Chat@b2c16d5842 (2026-08-18). Data as JSON: /api/errors/449189f0dec94b97. Report an issue: GitHub.