Summary

The chatbot loses all conversation history on pod restart because both storage layers default to ephemeral SQLite databases, despite the operator provisioning a PostgreSQL instance.

Lightspeed Stack generates its configmap without a database.postgres block, so the chatbot falls back to SQLite at /tmp/lightspeed-stack.db for the user_conversation and user_turn tables (sidebar metadata).

Llama Stack configures its kv_default and sql_default storage backends as kv_sqlite/sql_sqlite on the emptyDir volume, so the openai_conversations and conversation_items tables (actual message content) are also ephemeral.

Additionally, the chatbot_transcripts_enabled and chatbot_feedback_enabled settings are hardcoded to false in the configmap template with no CR spec field to override them.

Impact

• Conversation history disappears on every pod restart or page refresh
• The provisioned PostgreSQL database sits unused by the chatbot
• No supported way to enable file-based transcript/feedback exports

Expected Behavior

• The chatbot should use the existing PostgreSQL database (already provisioned by the postgres role) for all persistent conversation storage
• chatbot_transcripts_enabled and chatbot_feedback_enabled should be configurable via the CR spec

Steps to Reproduce

1. Deploy the operator and create an AnsibleAIConnect CR with a chatbot configured
2. Open the chatbot UI and send a message
3. Refresh the page or wait for a pod restart
4. Conversation history is gone
5. Inspect the lightspeed-stack configmap: no database.postgres block
6. Inspect the llama-stack configmap: sql_default uses sql_sqlite

Environment

• AAP 2.6 (operator builds 2.6.0+0.1774648973 and 2.6.0+0.1774648945)
• Observed on both OpenShift and EKS deployments

References

• lightspeed-core/lightspeed-stack database model
• Conversations API docs