Would automatic detection of stale MeiliSearch sync flags be a useful improvement?

[sruthi-sixt-26]

Hello LibreChat community, I am quite new to LibreChat and running it with MeiliSearch in Kubernetes without persistent storage for MeiliSearch.

After a MeiliSearch pod restart, I am hitting the documented scenario where LibreChat reports data as fully synced even though MeiliSearch is empty:

2026-03-11T23:13:07.372Z info: [indexSync] Conversations are fully synced: 500/500
2026-03-11T23:13:07.372Z debug: [MEILI_SYNC:meili-index-sync] Flow deleted
2026-03-11T23:13:07.372Z debug: [indexSync] Flow state cleaned up
2026-03-11T23:13:07.372Z warn: [FlowStateManager] Cannot complete flow - flow state not found
2026-03-11T23:13:07.372Z debug: [indexSync] No sync was needed

I understand from the MeiliSearch documentation that the reset-meili-sync script exists for exactly this case, and it works.

However, in a Kubernetes environment where MeiliSearch pods can be rescheduled or restarted at any time (without persistent volumes), this requires manual intervention each time.

The docs cover running the reset script for local and Docker setups, but not Kubernetes. For anyone else running in K8s, what is your go to approach? It would be great if the docs could include a Kubernetes section alongside the existing local and Docker instructions.

Idea

Would it be worthwhile for performSync() in indexSync.js to automatically detect this situation?

For example, comparing MeiliSearch's actual document count (via index.getStats()) against what MongoDB reports as synced. If MeiliSearch has 0 documents but MongoDB says "N" are synced, the _meiliIndex flags could be automatically reset essentially doing what reset-meili-sync does, but without manual intervention.

This would make LibreChat more resilient in container environments where MeiliSearch data loss is expected.

Questions for the community

• Has anyone implemented automatic detection for this in their deployment?
• Would this be a welcome PR, or is the current manual reset approach preferred to keep the sync logic simple?
• For those running in Kubernetes are you using persistent volumes for MeiliSearch, or do you have another strategy?

I also had a look into feat: add opensearch and typesense support, and looking forward to exploring it further

Thank you for the support!