VercelEngineering PostOfficial Source

Export AI Gateway traces with Vercel Drains

Vercel AI Gateway can export per-request OpenTelemetry traces, exposing routing, retries, latency, tokens, cost, and attribution without sending prompt or completion content.

Vercel · Aug 5, 2026

Open Source Open Markdown Open JSON

Source Summary

AI Gateway now emits **one OpenTelemetry trace per request**. Pro and Enterprise teams can export traces through Vercel Drains to OTLP/HTTP endpoints, with routing, fallbacks, retries, usage, cost, latency, status, and deployment attribution included.

Practical Implication

Send sampled traces to the observability system you already use, then inspect provider failover, slow first tokens, and unexplained agent costs alongside application telemetry. Prompt and completion content are excluded from the drain.

Agent-Ready Context

AI Gateway now emits **one OpenTelemetry trace per request**. Pro and Enterprise teams can export traces through Vercel Drains to OTLP/HTTP endpoints, with routing, fallbacks, retries, usage, cost, latency, status, and deployment attribution included.

Send sampled traces to the observability system you already use, then inspect provider failover, slow first tokens, and unexplained agent costs alongside application telemetry. Prompt and completion content are excluded from the drain.

Export costs **$0.05 per 1,000 delivered traces per drain**, plus **$0.50 per GB** transferred. Multiple provider attempts still count as one request, but sending the same request to multiple drains creates separate delivery charges.

Connected Context · Feed7 Judgment

This moves Gateway observability from a Vercel inspection surface into existing telemetry systems, preserving per-request routing, retry, cost, latency, status, and deployment context without exporting prompts or completions. It enables cross-stack diagnosis and routing analysis, while explicit delivery and transfer charges make sampling and drain count part of the observability design.

AI Gateway logs now have a dedicated pageDrains export much of the request-level evidence previously available on the dedicated logs page, enabling teams to correlate it with application telemetry and apply their own retention and alerting.AI Gateway now supports team and project spend budgetsTrace export supplies diagnostic evidence around costs and failures, while budgets provide enforcement; together they connect observation with a hard spending boundary that applications must handle.AI Gateway adds unified fast mode supportFast mode requires routing metadata to verify whether accelerated serving was used; exported traces make that verification and its latency-cost comparison available in external observability systems.TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AITRACE-Router argues for task-level routing based on final outcomes, while these traces provide request-level routing and performance evidence; additional task and outcome correlation would still be needed to evaluate that policy.

Context Map

infracoding#gateways#observability

Uncertainty

Export costs **$0.05 per 1,000 delivered traces per drain**, plus **$0.50 per GB** transferred. Multiple provider attempts still count as one request, but sending the same request to multiple drains creates separate delivery charges.