New trace-only mode: requests go directly to the LLM provider, bypassing the proxy hop. The SDK captures model, tokens, latency, status, and cost metadata asynchronously in the background. Same dashboard visibility, zero added latency.
Read the full post