diff --git a/packages/ai/package.json b/packages/ai/package.json
index 4c4fdb4c..089d86d6 100644
--- a/packages/ai/package.json
+++ b/packages/ai/package.json
@@ -82,8 +82,10 @@
     "c12": "^2.0.4",
     "commander": "^14.0.0",
     "defu": "^6.1.4",
+    "esbuild": "^0.25.8",
     "handlebars": "^4.7.8",
-    "nanoid": "^5.1.5"
+    "nanoid": "^5.1.5",
+    "vitest": "catalog:"
   },
   "peerDependencies": {
     "@opentelemetry/api": "^1.9.0",
@@ -105,13 +107,11 @@
     "@vitest/coverage-v8": "^3.2.4",
     "aiv4": "npm:ai@^4.3.19",
     "aiv5": "npm:ai@^5.0.0",
-    "esbuild": "^0.25.8",
     "eslint": "catalog:",
     "prettier": "catalog:",
     "tinyrainbow": "^2.0.0",
     "tsup": "catalog:",
     "typescript": "catalog:",
-    "vitest": "catalog:",
     "zod": "catalog:"
   },
   "files": [
diff --git a/packages/ai/test/evals/eval-integration.test.ts b/packages/ai/test/evals/eval-integration.test.ts
new file mode 100644
index 00000000..731d4474
--- /dev/null
+++ b/packages/ai/test/evals/eval-integration.test.ts
@@ -0,0 +1,228 @@
+// @vitest-environment node
+// @vitest-pool forks
+
+/**
+ * Integration test for Eval() that captures network calls and spans
+ */
+
+import { afterAll, vi } from 'vitest';
+import { InMemorySpanExporter, SimpleSpanProcessor } from '@opentelemetry/sdk-trace-base';
+import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
+import type { ResolvedAxiomConfig } from '../../src/config/index';
+import type { ReadableSpan } from '@opentelemetry/sdk-trace-base';
+
+const fetchCalls: Array<{ url: string; options: any }> = [];
+const consoleOutput: string[] = [];
+const spanExporter = new InMemorySpanExporter();
+
+const tracerProvider = new NodeTracerProvider({
+  spanProcessors: [new SimpleSpanProcessor(spanExporter)],
+});
+
+const mockConfig: ResolvedAxiomConfig = {
+  eval: {
+    url: 'https://test.axiom.co',
+    token: 'test-token',
+    dataset: 'test-dataset',
+    instrumentation: null,
+    include: [],
+    exclude: [],
+    timeoutMs: 60_000,
+  },
+} as ResolvedAxiomConfig;
+
+global.fetch = vi.fn(async (url: string, options?: any) => {
+  fetchCalls.push({ url: String(url), options });
+
+  if (url.includes('_apl')) {
+    return new Response(JSON.stringify({ matches: [] }), {
+      status: 200,
+      headers: { 'Content-Type': 'application/json' },
+    });
+  }
+
+  return new Response(JSON.stringify({ success: true }), {
+    status: 200,
+    headers: { 'Content-Type': 'application/json' },
+  });
+}) as any;
+
+const originalLog = console.log;
+console.log = (...args: any[]) => {
+  consoleOutput.push(args.map(String).join(' '));
+};
+
+vi.doMock('../../src/evals/instrument', async () => {
+  const { trace: _trace } = await import('@opentelemetry/api');
+  const tracer = tracerProvider.getTracer('axiom-eval-test');
+
+  return {
+    ensureInstrumentationInitialized: vi.fn(async () => {}),
+    initInstrumentation: vi.fn(async () => {}),
+    flush: vi.fn(async () => {
+      await tracerProvider.forceFlush();
+    }),
+    startSpan: vi.fn((name: string, opts: any, ctx?: any) => {
+      return tracer.startSpan(name, opts, ctx);
+    }),
+    startActiveSpan: vi.fn(async (name: string, opts: any, fn: any, ctx?: any) => {
+      const span = tracer.startSpan(name, opts, ctx);
+      try {
+        const result = await fn(span);
+        span.end();
+        return result;
+      } catch (error) {
+        span.recordException(error as Error);
+        span.end();
+        throw error;
+      }
+    }),
+  };
+});
+
+// Mock context storage
+vi.doMock('../../src/evals/context/storage', () => ({
+  getAxiomConfig: vi.fn(() => mockConfig),
+  setAxiomConfig: vi.fn(),
+  getConfigScope: vi.fn(() => ({
+    getAllDefaultFlags: () => ({}),
+  })),
+  withEvalContext: vi.fn(async (_opts: any, fn: any) => {
+    const result = await fn();
+    return result;
+  }),
+  getEvalContext: vi.fn(() => ({
+    flags: {},
+    outOfScopeFlags: [],
+  })),
+}));
+
+// Mock global flags
+vi.doMock('../../src/evals/context/global-flags', () => ({
+  getGlobalFlagOverrides: vi.fn(() => ({})),
+  setGlobalFlagOverrides: vi.fn(),
+}));
+
+// Mock vitest inject() to provide context
+vi.doMock('vitest', async () => {
+  const actual = await vi.importActual('vitest');
+  return {
+    ...actual,
+    inject: vi.fn((key: string) => {
+      const context: Record<string, any> = {
+        baseline: undefined,
+        debug: false,
+        list: false,
+        overrides: {},
+        axiomConfig: mockConfig,
+        runId: 'test-run-123',
+      };
+      return context[key];
+    }),
+  };
+});
+
+const { Eval } = await import('../../src/evals/eval');
+const { createScorer: Scorer } = await import('../../src/evals/scorers');
+
+const testScorer = Scorer('test-scorer', async ({ output }: { output: any }) => {
+  return {
+    score: typeof output === 'string' && output.includes('output') ? 1.0 : 0.0,
+  };
+});
+
+const scorer1 = Scorer('scorer-1', async ({ output: _output }: { output: any }) => ({
+  score: 1.0,
+}));
+
+const scorer2 = Scorer('scorer-2', async ({ output: _output }: { output: any }) => ({
+  score: 0.8,
+}));
+
+Eval('Integration-Test-Eval', {
+  data: async () => [
+    { input: 'test input 1', expected: 'expected output 1' },
+    { input: 'test input 2', expected: 'expected output 2' },
+  ],
+  task: async ({ input }) => {
+    return `output for ${input}`;
+  },
+  scorers: [testScorer as any],
+});
+
+Eval('Second-Eval', {
+  data: async () => [{ input: 'input A', expected: 'expected A' }],
+  task: async ({ input }) => `output for ${input}`,
+  scorers: [scorer1 as any],
+});
+
+Eval('Third-Eval', {
+  data: async () => [
+    { input: 'input X', expected: 'expected X' },
+    { input: 'input Y', expected: 'expected Y' },
+  ],
+  task: async ({ input }) => `result for ${input}`,
+  scorers: [scorer2 as any],
+});
+
+afterAll(async () => {
+  console.log = originalLog;
+
+  const spans: ReadableSpan[] = spanExporter.getFinishedSpans();
+
+  const baselineCall = fetchCalls.find((c) => c.url.includes('_apl'));
+  const createCall = fetchCalls.find(
+    (c) => c.url.includes('/api/evaluations/v3') && c.options?.method === 'POST',
+  );
+  const updateCall = fetchCalls.find(
+    (c) => c.url.includes('/api/evaluations/v3') && c.options?.method === 'PATCH',
+  );
+
+  if (!baselineCall) throw new Error('Expected baseline query call');
+  if (!createCall) throw new Error('Expected create evaluation call');
+  if (!updateCall) throw new Error('Expected update evaluation call');
+
+  // Assert span structure for Integration-Test-Eval (comprehensive)
+  const evalSpan = spans.find((s) => s.name.includes('eval Integration-Test-Eval'));
+  const allCaseSpans = spans.filter((s) => s.name.startsWith('case'));
+  const integrationCaseSpans = allCaseSpans.slice(0, 2); // First 2 cases are from Integration-Test-Eval
+  const taskSpans = spans.filter((s) => s.name === 'task');
+  const scorerSpans = spans.filter((s) => s.name.includes('score'));
+
+  if (!evalSpan) throw new Error('Expected eval span');
+  if (integrationCaseSpans.length !== 2) {
+    throw new Error(
+      `Expected 2 Integration-Test-Eval case spans, got ${integrationCaseSpans.length}`,
+    );
+  }
+  if (taskSpans.length < 2) {
+    throw new Error(`Expected at least 2 task spans, got ${taskSpans.length}`);
+  }
+  if (scorerSpans.length < 2) {
+    throw new Error(`Expected at least 2 scorer spans, got ${scorerSpans.length}`);
+  }
+
+  // Assert span attributes
+  const firstCaseSpan = integrationCaseSpans[0];
+  const attrs = firstCaseSpan.attributes;
+
+  if (!attrs['eval.case.input']) throw new Error('Expected eval.case.input attribute');
+  if (!attrs['eval.case.output']) throw new Error('Expected eval.case.output attribute');
+  if (!attrs['eval.case.scores']) throw new Error('Expected eval.case.scores attribute');
+
+  // Assert multiple evals ran (light validation)
+  const secondEvalSpan = spans.find((s) => s.name.includes('Second-Eval'));
+  const thirdEvalSpan = spans.find((s) => s.name.includes('Third-Eval'));
+
+  if (!secondEvalSpan) throw new Error('Expected Second-Eval span');
+  if (!thirdEvalSpan) throw new Error('Expected Third-Eval span');
+
+  // Total case count should be 2 + 1 + 2 = 5
+  if (allCaseSpans.length !== 5) {
+    throw new Error(`Expected 5 total case spans, got ${allCaseSpans.length}`);
+  }
+
+  // Cleanup
+  await tracerProvider.shutdown();
+  await spanExporter.shutdown();
+});
diff --git a/packages/ai/vitest.config.ts b/packages/ai/vitest.config.ts
index 887054f4..8f3512af 100644
--- a/packages/ai/vitest.config.ts
+++ b/packages/ai/vitest.config.ts
@@ -5,6 +5,7 @@ export default defineConfig({
   test: {
     environment: 'node',
     include: ['test/**/*.test.ts'],
+    exclude: ['test/evals/**/*.integration.test.ts', '**/node_modules/**'],
     globals: true,
     pool: 'forks',
     // TODO: ensure that this allows parallel tests