Vercel Brings AI Gateway Trace Exports to Vercel Drains.

Vercel is opening up the telemetry generated by its AI infrastructure. As of August 5, 2026, teams can export AI Gateway traces through Vercel Drains, sending request-level data to external observability services instead of keeping their debugging workflow confined to Vercel.

The change gives developers a more direct way to investigate model requests, monitor production AI applications, and combine AI telemetry with data from the rest of their stack. According to the official changelog, AI Gateway now creates an OpenTelemetry trace for every request.

Pro and Enterprise teams can route those traces to any endpoint compatible with OpenTelemetry Protocol over HTTP, commonly referred to as OTLP/HTTP. Vercel also provides native integrations for Braintrust, Dash0, Kubiks, Sentry, and Statsig.

AI applications are often harder to debug than conventional web services. A successful HTTP response does not necessarily mean the underlying workflow behaved correctly. Model latency, provider selection, tool calls, retries, and downstream failures can all affect what the user ultimately sees.

Exportable traces give teams a common mechanism for examining those requests alongside the rest of their application telemetry. For a small business operating an AI assistant, for example, this could make it easier to determine whether slow responses originate in the application, the gateway, or an external model provider. Teams already using an observability platform can keep that investigation in an existing dashboard rather than adding another isolated monitoring surface.

The use of OpenTelemetry is important. It is an established, vendor-neutral framework for producing and transporting telemetry, so Vercel customers are not limited to the named integrations. Any service capable of receiving traces through an OTLP/HTTP endpoint can potentially serve as the destination.

The practical setup includes three controls highlighted by Vercel:

• Selecting an external destination for the trace drain
• Choosing which projects should send traces
• Setting a sampling rate to control how much request data is exported

Sampling can be particularly relevant for applications handling substantial traffic. Exporting every trace provides the fullest debugging record, but a lower sampling rate may be more appropriate when teams need to manage observability volume and destination-side costs. Vercel did not announce separate pricing for the feature in the changelog.

The update builds on Vercel’s broader use of drains as an integration layer between its platform and third-party monitoring systems. The company has also expanded Audit Log Drains, including support for external security and analytics services, as Create With previously reported.

AI Gateway trace exports apply that same general pattern to model traffic. Instead of requiring teams to manually instrument every gateway request or build a custom export pipeline, Vercel produces the OpenTelemetry trace and forwards it through the configured drain.

That matters as AI Gateway expands beyond basic model routing. Vercel has continued adding models and agent integrations, including Grok Imagine Image 2.0 support and connectivity with Hermes Agent and Vercel Sandbox. More model choices and agentic workflows can create more points of failure, increasing the value of centralized traces.

There are limits to what Vercel disclosed. The short announcement does not enumerate every field included in each trace, specify retention behavior at third-party destinations, or detail destination-specific configuration. Teams handling sensitive prompts or outputs should therefore review the trace contents and the data policies of their chosen observability provider before enabling broad exports.

For eligible teams, however, the core benefit is straightforward: AI Gateway activity can now participate in the same observability workflow as the rest of a production application.

What is Vercel Drains?

Vercel Drains is a mechanism for sending telemetry generated by Vercel to external systems. With this update, it can export OpenTelemetry traces produced for AI Gateway requests to compatible observability endpoints.

Which plans can export AI Gateway traces?

Vercel says the capability is available to Pro and Enterprise teams. The announcement does not provide separate pricing for trace exports or describe availability on other plans.

Which observability platforms are supported?

Vercel lists native integrations for Braintrust, Dash0, Kubiks, Sentry, and Statsig. Teams can also send traces to any external destination that accepts OTLP over HTTP.