@veryfront/ext-eval-report-http

Category: Eval export | Requires: EvalReportExporterRegistry | Optional

Registers HTTP-backed eval report exporters. Use this extension when a project needs to send redacted EvalReport payloads to an internal endpoint, Braintrust, Langfuse, LangSmith, or another eval platform through a gateway.

The extension does not own runtime tracing. Use @veryfront/ext-observability-opentelemetry for OpenTelemetry spans, metrics, and service monitoring.

Eval report HTTP export is an explicit data export path. It sends the completed, redacted EvalReport and its export context to the configured endpoint only when an eval run selects the exporter. Use this extension for Langfuse, LangSmith, Braintrust, or internal gateway integrations that need eval records, scores, redaction policy, and optional trace correlation. Do not use OTLP runtime telemetry env vars to route eval reports; OTEL_* settings only control runtime trace and metric export.

Installation

Add the extension to your project's veryfront.config.ts:

import extEvalReportHttp from "@veryfront/ext-eval-report-http";

export default defineConfig({
  extensions: [
    extEvalReportHttp({
      exporters: [
        {
          id: "braintrust-proxy",
          url: "https://evals.example.com/reports",
          token: "<TOKEN>",
          headers: { "x-workspace": "default" },
        },
      ],
    }),
  ],
});

Then run an eval with a matching exporter id:

veryfront eval deep-research --export braintrust-proxy

Environment variables

Without factory configuration, the extension registers one exporter from env configuration.

| Variable | Required | Description | | -------------------------------------- | -------- | ------------------------------------------------- | | VERYFRONT_EVAL_HTTP_EXPORTER_URL | Yes | HTTP endpoint that receives { report, context } | | VERYFRONT_EVAL_HTTP_EXPORTER_ID | No | Exporter id. Defaults to http | | VERYFRONT_EVAL_HTTP_EXPORTER_TOKEN | No | Bearer token for the authorization header | | VERYFRONT_EVAL_HTTP_EXPORTER_HEADERS | No | JSON object or comma-separated key=value pairs |

Factory configuration

extEvalReportHttp({
  exporters: [
    {
      id: "langfuse-proxy",
      url: "https://evals.example.com/langfuse",
      token: "<TOKEN>",
      headers: { "x-project": "docs-agent" },
      method: "POST",
    },
  ],
});

Each exporter sends:

{
  "report": {},
  "context": {}
}

runEval enriches export context with the active runtime traceId and spanId when OpenTelemetry is active and the caller did not pass an explicit context.trace.

That trace context is correlation metadata only. It does not include span data, metric streams, or logs, and changing the eval HTTP exporter does not change the OpenTelemetry runtime exporter.

Required contract

EvalReportExporterRegistry is seeded by Veryfront bootstrap. The extension requires that registry during setup and registers one EvalReportExporter per configured endpoint. Teardown unregisters only the exporter ids that this extension registered.

Capabilities

• net *: sends eval reports to the configured endpoint.
• env: reads the four VERYFRONT_EVAL_HTTP_EXPORTER_* variables listed above when explicit factory configuration is not set.