Problem
Ask ALYF Conversation.messages_json stores only the user-visible turns. The internal tool round-trips, per-turn model latency and token usage are not persisted, which makes performance diagnosis hard (e.g. measuring how many round-trips a question actually took).
Proposal
Persist per-turn telemetry — number of internal tool round-trips, model latency, and token usage — either in Ask ALYF Conversation or a dedicated log DocType.
Benefit
Makes latency/cost regressions measurable and supports tuning (e.g. validating prompt-caching or context-capping changes).
Context
Feedback from production use with ERPNext v15 + Aakvatech PropMS.
Problem
Ask ALYF Conversation.messages_jsonstores only the user-visible turns. The internal tool round-trips, per-turn model latency and token usage are not persisted, which makes performance diagnosis hard (e.g. measuring how many round-trips a question actually took).Proposal
Persist per-turn telemetry — number of internal tool round-trips, model latency, and token usage — either in
Ask ALYF Conversationor a dedicated log DocType.Benefit
Makes latency/cost regressions measurable and supports tuning (e.g. validating prompt-caching or context-capping changes).
Context
Feedback from production use with ERPNext v15 + Aakvatech PropMS.