agtrace-provider-normalization

To unify AI provider logs, implement a three-tier adapter pattern consisting of trait-based adapters, provider-specific implementations, and unified domain types. This normalizes diverse log formats into a single domain model.

Schema-on-read normalizes AI log formats by treating raw logs as the source of truth and applying lazy parsing to achieve type-safe conversion. This defers structural parsing until the data is queried or processed.

To add a new provider, implement Discovery, Parser, and ToolMapper components, then register the adapter in the provider registry. This wires the new provider into the existing unified domain type architecture.

Tool calls are normalized across AI agent providers using a ToolMapper component within the adapter architecture. It translates provider-specific tool invocation formats into unified domain types.

You need multi-stream events with deterministic UUIDs when building unified events from asynchronous AI log streams. This ensures that related log entries across different providers can be deterministically correlated.